OECD PRODUCTIVITY WORKING PAPERSJanuary 2019 No. 17

UNCONVENTIONAL MONETARY POLICY AND PRODUCTIVITY: EVIDENCE ON THE RISK-SEEKING CHANNEL FROM US CORPORATE BOND MARKETSSILVIA ALBRIZIOMARINA CONESADENNIS DLUGOSCHCHRISTINA TIMILIOTIS

│ 1

OECD PRODUCTIVITY WORKING PAPERS

The OECD Productivity Working Papers are associated with the Global Forum on Productivity that provides a forum for mutual exchange of information and fosters international co-operation between public bodies with responsibility for promoting productivity-enhancing policies, including in undertaking joint policy analysis. It offers a platform for exchanging views, experiences and information, institutional and governance arrangements and government structures, with a view towards developing better policies. The Forum extends existing work in the OECD through a well-prioritised and coherent stream of analytical work serving the policy research needs of participants on the drivers of productivity growth.

This paper has been authorised for publication by Luiz de Mello, Director, Policy Studies Branch.

OECD Productivity Working Papers should not be reported as representing the official views of the OECD or of its member countries. The opinions expressed and arguments employed are those of the author(s).

Comments on OECD Productivity Working Papers are welcome and may be sent to: productivity@oecd.org

OECD Productivity Working Papers are published on: oe.cd/productivity

Series: OECD Productivity Working Paper Series

ISSN 2413-9424

Please cite this paper as:

Albrizio, S., Conesa, M., Dlugosch, D. and Timiliotis, C., “Unconventional monetary policy and productivity: Evidence on the risk-seeking channel from US corporate bond markets”, OECD Productivity Working Papers, 2019-17, OECD Publishing, Paris.

This document and any map included herein are without prejudice to the status of or sovereignty over any territory, to the delimitation of international frontiers and boundaries and to the name of any territory, city or area.

© OECD (2019)

You can copy, download or print OECD content for your own use, and you can include

excerpts from OECD publications, databases and multimedia products in your own

documents, presentations, blogs, websites and teaching materials, provided that suitable

acknowledgment of OECD as source and copyright owner is given. All requests for

commercial use and translation rights should be submitted to rights@oecd.org.

2 │

ABSTRACT/RÉSUMÉ

Unconventional monetary policy and productivity: Evidence on the risk-seeking channel

from US corporate bond markets

We examine the relationship between lax monetary policy, access to high-yield bond markets

and productivity in the US between 2008 and 2016. Using monetary policy surprises, obtained

from changes in interest rates futures in narrow windows around FOMC announcements, we

isolate the increased access to high-yield bond markets relative to investment-grade bond markets

that is due to unconventional monetary policy (UMP). We find that through the risk-taking

channel, UMP has increased investors’ appetite for high-yield US corporate bonds, thereby

increasing access to high-yield bond markets for firms with a higher risk profile. Since the

relationship between credit ratings and firm-level productivity is U-shaped, the aggregate effect

on productivity is a priori unclear. Turning to the real economy, we thus analyse whether this

additional access to finance had an effect on aggregate productivity by altering the reallocation

of resources across firms. Our results show that unconventional monetary policy induced less

investment in tangible capital by high-productive firms. However, before drawing conclusions

on the net effects of UMP on aggregate productivity, we discuss a number of issues that this

paper could not deal with due to data limitations, including prominently whether this apparent

misallocation may have been offset by a shift in the composition of investments towards more

intangible investment

JEL classification: D24, E52, G21, G32, G33, J63, O16 , O47.

Keywords: Unconventional monetary policy, productivity, capital reallocation, United States,

bond markets.

****************

Politique monétaire non conventionnelle et productivité : observation du canal de la prise

de risque sur les marchés des obligations d’entreprises aux États-Unis

Nous examinons la relation entre une politique monétaire laxiste, l’accès aux marchés

obligataires à haut rendement et la productivité aux États-Unis entre 2008 et 2016. À partir des

effets de surprise de la politique monétaire, calculés comme correspondant aux variations de taux

sur les contrats à terme de taux d’intérêt sur les courts laps de temps entourant les dates des

annonces du Comité de politique monétaire de la Réserve fédérale (FOMC), nous mettons en

évidence un accès aux marchés obligataires à haut rendement élargi par rapport à l’accès aux

marchés d’obligations de qualité d’investissement, qui est dû à la politique monétaire non

conventionnelle. Nous constatons en effet que par le canal de la prise de risque, la politique

monétaire non conventionnelle a accru l’appétence des investisseurs pour les marchés

d’obligations d’entreprises à haut rendement aux États-Unis, permettant ainsi aux entreprises à

profil plus risqué d’accéder plus largement aux marchés obligataires à haut rendement. Étant

donné le profil en U de la courbe entre la note de crédit et la productivité au niveau des

entreprises, l’effet global sur la productivité est a priori peu évident. Nous attachant ensuite aux

effets sur l’économie réelle, nous examinons si cet élargissement de l’accès au financement a eu

un effet sur la productivité globale par un redéploiement des ressources entre les entreprises. Nos

résultats montrent que la politique monétaire non conventionnelle a induit moins

d’investissements dans les actifs corporels de la part des entreprises très productives. Cela étant,

avant de tirer des conclusions sur l’impact net de la politique monétaire non conventionnelle sur

la productivité globale, nous examinons un certain nombre de questions que le présent article n’a

pu aborder, faute de données disponibles, au premier rang desquelles la question de savoir si cette

mauvaise affectation apparente des ressources peut avoir été compensée par un changement de

composition de l’investissement au profit d’investissements plus importants dans les actifs

incorporels.

Classification JEL : D24, E52, G21, G32, G33, J63, O16, O47

Mots-clés : Politique monétaire non conventionnelle, productivité, réaffectation des ressources,

États-Unis, marchés obligataires.

│ 3

Table of contents

Unconventional Monetary Policy and productivity: Evidence on the risk-seeking channel from

US corporate bond markets .................................................................................................................. 5

1. Introduction ...................................................................................................................................... 5 2. The evolution of corporate bond markets in the US and its relationship with (unconventional)

monetary policy ................................................................................................................................... 8 3. Related literature and the economics behind the risk-seeking channel ............................................ 9

3.1. The broader finance-productivity link ..................................................................................... 10 3.2. The real economy effects of unconventional monetary policy ............................................... 10 3.3. The economics of the risk-seeking channel ............................................................................ 10

4. Data ................................................................................................................................................ 11 4.1. Bond issuance data .................................................................................................................. 11 4.2. Balance-sheet data ................................................................................................................... 12 4.3. Matching ................................................................................................................................. 13 4.4. Productivity measurement ....................................................................................................... 13

5. Empirical approach ........................................................................................................................ 14 5.1. The proxy SVAR framework .................................................................................................. 14 5.2. VAR Estimation and the choice of the monetary policy indicator and instrument ................. 16 5.3. Monetary policy shocks and capital reallocation .................................................................... 17

6. Results ............................................................................................................................................ 20 6.1. Did expansionary monetary policy increase demand-driven risk-seeking behaviour in

corporate bond markets? ................................................................................................................ 20 6.2. Did monetary policy shocks distort the capital reallocation process? ..................................... 24 6.3. Robustness checks ................................................................................................................... 28 6.4. Discussion and implications for further research .................................................................... 32

7. Conclusion and policy implications ............................................................................................... 35 7.1. Conclusion ............................................................................................................................... 35 7.2. Policy implications .................................................................................................................. 36

References ............................................................................................................................................. 37

Annex A. Data ...................................................................................................................................... 41

Cleaning and matching ................................................................................................................... 41 Filtering and Cleaning .................................................................................................................... 41

Annex B. Robustness and additional results ..................................................................................... 43

Tables

Table 1. Structural identification: First-stage results ............................................................................. 21 Table 2. Reallocation - Baseline results ................................................................................................ 25

Table A.1. Summary statistic ................................................................................................................ 41

4 │

Figures

Figure 1. US corporate bond issuance: investment grade and high yield (2008=100) ............................ 6 Figure 2. Corporate bond ratings and the distribution of productivity .................................................... 6 Figure 3. Unconventional monetary policy and the corporate bond market ........................................... 9 Figure 4. The ratio of HY/IG due to monetary policy shock................................................................. 22 Figure 5. Risk-seeking in corporate bond markets after an unconventional expansionary monetary

policy shock ................................................................................................................................... 24 Figure 6. Marginal effects of increased bond issuance due to monetary policy surprises on firm-level

capital expenditures ....................................................................................................................... 26 Figure 7. Marginal effects of increasing MFP on the propensity to issue HY bonds ............................ 27 Figure 8. The use of HY corporate bonds by MFP quartile .................................................................. 33

Figure B.1. Impulse-response functions after an expansionary monetary policy shock ....................... 43 Figure B.2. Different VAR estimation window: Impulse-response functions after an expansionary

monetary policy shock ................................................................................................................... 45

│ 5

Unconventional Monetary Policy and productivity: Evidence on the risk-

seeking channel from US corporate bond markets

By Silvia Albrizio, Marina Conesa, Dennis Dlugosch and Christina Timiliotis1

1. Introduction

1. The productivity slowdown the US has experienced since mid-2000s has

triggered an intense debate among policymakers on the underlying causes without

reaching a unique consensus (Gordon, 2012; Summers, 2016, Byrne et al., 2016). Within

this debate, policymakers and academics’ attention has only recently been drawn to the

unintended consequences that macroeconomic policies might have had on productivity

by meddling with the resource and “creative destruction” process of market economies

(Obstfeld, 2018; Borio, 2018; Adalet McGowan et al., 2017a; 2017b).

2. Monetary policy constitutes an obvious candidate in this context given the

relatively lax stance adopted by many advanced economies in this regard since the early

2000s, and notably following the financial crisis. As the world faced its worst economic

crisis since the Great Depression, major central banks including the Fed, indeed

introduced record-low short-term interest rates in an effort to promote economic growth.

Conventional monetary policies became limited in their effectiveness, however, as

short-term interest rates approached the zero lower bound. The Fed and other major

central banks thus reverted to the use of unconventional monetary policies, including a

series of large-scale asset purchases. Taken together, these measures significantly

contributed to flattening the yield curve, and are likely to have fuelled investors’ risk-

taking behaviour (Borio and Zhu, 2008; Lhuissier and Szczerbowicz, 2017).

3. One of the channels through which investors’ risk-taking could have

materialized is the search-for-yield. This behaviour could unfold in a situation where

investors have sticky return targets, but are unable to meet these due to low interest rates.

This would then raise their incentive to take on more risk in order to meet future

obligations. Examples of investors to which this applies notably include institutional

investors, such as pension funds and insurance companies.

4. High-yield bonds, as opposed to government bonds, provide one way to boost

the expected return profile at the cost of being more risky. Figure 1 shows that the

demand for corporate bonds of this type has drastically increased over the past decade

In particular, this increase in the supply of high-yield corporate bonds coincided with

the scale-up of quantitative easing by the Fed and possibly opened up the bond market

to firms with a high risk profile that may not have received financing otherwise.

Implications on the real economy and on aggregate productivity of this increase in

1 Corresponding authors are: Silvia Albrizio (Bank of Spain; silvia.albrizio@bde.es), Marina

Conesa (Inter-American Development Bank; marinaco@iadb.org), Dennis Dlugosch (OECD

Economics Department; Dennis.Dlugosch@oecd.org ) and Christina Timiliotis (OECD

Economics Department; Christina.Timiliotis@oecd.org ). The authors are grateful to Giuseppe

Nicoletti, Douglas Sutherland, Serdar Celik, Stéphane Sorbe, Peter Gal (from the OECD

Economics Department), Ángel Estrada, Luna Romo, Gonzalez, Davide Furceri, Omar Rachedi

(from Bank of Spain), Peter Karadi (ECB), Romain Duval (IMF), and delegates to the OECD

Global Forum on Productivity for their valuable comments. Further thanks go to Sarah Michelson

(OECD Economics Department) for excellent editorial support and to Emilio Muñoz de la Peña

(Bank of Spain) for excellent technical support.

6 │

corporate bond issuance are a priori unclear, insofar as the relationship between credit

ratings and productivity is non-linear, as shown by Figure 2, and several recent

contributions to the literature (see for instance Aghion et al., 2018; Bakhtiari, 2017).

Accordingly, the same low credit rating could mask two different types of high-

leveraged firms: (i) low-productive companies with low profits, repayment capacity and

investment (“zombies”), or (ii) high-productive firms with unproven operating histories,

high borrowing and investment (“gazelles”). In sum, credit ratings thus cannot help to

understand whether the increase in access to high-yield bond markets due to the risk-

taking channel of monetary policy eventually facilitates or impedes a growth-improving

reallocation of resources.

Figure 1. US corporate bond issuance: investment grade and high yield (2008=100)

Note: Data refer to non-convertible bonds of non-financial companies.

Source: Dealogic

Figure 2. Corporate bond ratings and the distribution of productivity

Note: Credit ratings are defined on a range from 21=AAA to 1=CC. We consider average credit ratings and

multifactor-productivity performance by firm over the period 2008-16.The construction of MFP values is

based on the Solow measure (see Gal, 2013). Data refers to consolidated data for US non-financial firms,

only.

Source: Calculations based on Thomson Reuters Worldscope and Dealogic.

0

50

100

150

200

250

300

350

400

2008 2009 2010 2011 2012 2013 2014 2015 2016

High Yield Investment Grade

│ 7

5. This paper sheds light on the empirical relationship between unconventional

monetary policy and capital reallocation in the US using a two-stage approach. First, we

assess to what extent the recent growth in supply of high-yield bonds relative to

investment-grade bonds can be attributed to monetary policy shocks. Second, we test if

the increased access to high-yield bond markets induced by expansionary monetary

policy negatively weighed on capital reallocation, by allocating capital to low instead of

high-productive firms. Given that the misallocation of resources has been associated

with important losses in aggregate productivity (Restuccia and Rogerson, 2008; Hsieh

and Klenow, 2009), this finding would bear implications for the productivity slowdown

debate.

6. The analysis is developed mirroring these two steps. Following the approach

pioneered by Gertler and Karadi (2015), we identify shocks to monetary policy using

surprises in interest rate futures on Treasury bonds. Surprises are computed as changes

in interest rates futures in narrow windows around Federal Open Market Committees

(FOMC) meetings, and are used to instrument the indicators of monetary policy (short

and long interest rates of government bonds). Using a monthly structural VAR

framework, we then identify the variation in the ratio of high-yield over investment

grade gross bond issuance that can be attributed to monetary policy. Since higher

demand for HY relative to IG bonds is expected to drive up the prices for HY bonds and

decrease their yields, we define risk-taking as a simultaneous increase in the relative

high-yield bond issuance and decrease in the spread. Contrary to conventional monetary

policy which mainly acts on short term interest rates, UMPs are introduced with the

ultimate aim of driving down longer term interest rates (Krishnamurthy and Vissing-

Jorgensen, 2011). Our approach thus identifies shocks to unconventional monetary

policies (UMP) by using interest rates on Treasury bonds with long maturities as

monetary policy indicators.

7. Following the identification of UMP shocks, we use the estimated relative

policy-induced increase in high-yield bond issuance (HY/IG) to assess its effect on

reallocation. In line with canonical models of firm dynamics, we measure reallocation

as the covariance between firm productivity and growth, captured by capital

accumulation (see Foster et al., 2016; Decker et al., 2016). The intuition behind this

model is that firms with a higher productivity should be able to attract more (financial)

resources and grow faster. To test for the potentially distortionary effect of HY/IG

shocks on resource allocation, we then augment this framework with an interaction

between productivity and the relative change in high-yield bond issuance due to UMP

shocks. Results show that unconventional monetary policy have favoured investment in

tangible capital in the least productive firms, allowing them to accumulate fixed capital

and grow disproportionally more than more productive ones. We further show that this

heterogeneous effect on investment might result from the intrinsically higher propensity

to issue HY bonds of less productive firms, pointing towards a potential crowding-out

effect. A priori, the main finding from the reallocation model would point to a

misallocation of resources. However, before drawing conclusions on the net effects on

aggregate productivity, further research is needed to explore a number of issues that this

paper could not deal with due to data limitations, including prominently whether this

apparent misallocation may have been offset by the use of HY bond proceeds to finance

productive intangible investment or other uses of proceeds in the most productive firms.

8. The remainder of this paper is organized as follows. Section 2 provides a brief

overview of the evolution of corporate bond markets in the US since the 2000s and their

relation to monetary policy. Section 3 further explains the monetary policy-productivity

8 │

channel explored in the paper and surveys the related literature; section 4 discusses our

dataset; section 5 illustrates our empirical strategy; section 6 presents and discusses the

empirical findings; and section 7 concludes.

2. The evolution of corporate bond markets in the US and its relationship with

(unconventional) monetary policy

9. The extraordinary turmoil that hit global financial markets during the great

financial crisis and the subsequently sluggish pace of economic recovery forced central

banks around the world, including the FED, to take a number of unconventional steps to

alleviate financial distress and support economic activity. Once the target federal funds

rate was lowered to its effective zero lower bound towards the end of 2008 (Figure 3,

Panel A), limiting the scope for further conventional monetary policy, the FED began

to increase its monetary basis by purchasing large quantities of government debt

securities in the secondary market (Figure 3, Panel B).2 Both in terms of scale and

prominence, this step was considered unprecedented (Gilchrist and Zakrajsek, 2013).

10. The above mentioned large-scale asset purchase program (LSAP), which made

headlines under the name of “Quantitative Easing”, involved the purchasing of (i) debt

obligations of the government-sponsored housing agencies (GSEs), (ii) mortgage-

backed securities (MBS) issued by those agencies, and (iii) coupon securities issued by

the Treasury. Contrary to conventional monetary policy tools, these programs were

targeted at long-term assets, with a view to lowering long-term interest rates, which had

remained largely unaffected. Besides the launch of the three LSAP programs over the

period 2008-12 (QE1, QE2, QE3), the FED later also introduced the Maturity Extension

Program (MEP) that further extended the average maturity of its holdings of securities,

again, with the aim of putting further downward pressure on long-term interest rates (see

Fawley and Neely, 2013, for a more detailed description of these programs).

11. Further to historically low interest rates, debt security markets synchronously

endured a steady retrenchment in bank lending. Many banks struggled to cope with the

legacy of the crisis, weakened by trading losses and credit provisions. On top of this,

stricter prudential regulations and higher capital requirement, as well as a more prudent

stance towards risky borrowers, trimmed their lending capacity. Together, these

circumstances contributed to creating an environment that boosted the demand for

corporate debt securities, promising relatively higher returns, thereby boosting the

attractiveness of market debt as a further financing option for non-financial corporations.

12. While the deepening of bond markets is per se beneficial for firms, considering

that it increases funding diversification and lengthens debt maturity, the post-crisis surge

in bond finance has been accompanied by a decrease in credit quality, as evidenced by

the rise in the share of high-yield corporate bonds (cf. Figure 1). The real effects of this

development have not yet been gauged, but there are good reasons to believe that debt

markets may have become more fragile by favouring firms with low growth potential,

thus exposing bond holders to greater risks.

2 Unconventional monetary policy tools further include forward guidance – the verbal assurance of central

banks on their intended monetary policy actions.

│ 9

Figure 3. Unconventional monetary policy and the corporate bond market

Panel A: The inverse relation between corporate bond issuance and the Federal Funds Rate

Panel B: US Treasury Securities held by the Federal Reserve (all maturities)

Note: Federal Funds Rate data are based on monthly, not seasonally adjusted data. Corporate bond issuance

data cover non-convertible high-yield and investment grade bonds (Panel A). Total face value of US

Treasury securities held by the Federal Reserve (Panel B).

Source: Federal Reserve Bank of St. Louis; Securities Industry and Financial Markets Association (SIFMA)

based on Thomson Reuters

3. Related literature and the economics behind the risk-seeking channel

13. Given the complexity of the transmission channel analyzed hereafter, which

involves monetary policy, financial instruments and resource reallocation, the paper

relates to several strands of literature. To our best knowledge, no single paper so far

assessed all three factors in conjunction.

0

1

2

3

4

5

6

7

0

200

400

600

800

1000

1200

1400

1600

1800

19

99

-11

-01

20

00

-08

-01

20

01

-05

-01

20

02

-02

-01

20

02

-11

-01

20

03

-08

-01

20

04

-05

-01

20

05

-02

-01

20

05

-11

-01

20

06

-08

-01

20

07

-05

-01

20

08

-02

-01

20

08

-11

-01

20

09

-08

-01

20

10

-05

-01

20

11

-02

-01

20

11

-11

-01

20

12

-08

-01

20

13

-05

-01

20

14

-02

-01

20

14

-11

-01

20

15

-08

-01

20

16

-05

-01

20

17

-02

-01

20

17

-11

-01

20

18

-08

-01

IG HY FFR

Federal Funds RateUS Corporate Bond Issuance (USD

0

500

1000

1500

2000

2500

3000

BIL

LIO

NS

OF

DO

LLA

RS

Start of QE-1(25 Nov 2008)

Start of QE-2(10 Aug 2010)

Start of MEP(21 Sep 2011)

Start of QE-3(22 Aug 2012)

10 │

3.1. The broader finance-productivity link

14. This paper adds to a body of research investigating the role of financial factors

on productivity. Empirical evidence generally suggests that financial frictions tend to

reduce aggregate productivity and innovative activity (Buera et al., 2011; Duval et al.

2017; Heil, 2017; Gopinath et al. 2018; Demmou et al, 2018), including economy’s

churn rates, i.e. the sum of births and deaths of enterprises, one critical indicator of a

well-functioning reallocation process (Adalet McGowan et al. 2017a, 2017b; Andrews

and Petroulakis, 2017). Our paper also relates to the literature on the effects of

differences in capital structures on firm-level investment and innovation where the

empirical evidence shows that in countries which rely more on debt- than equity

financing, the level of innovation tends to be significantly lower (Kortum and Lerner,

1998; Brown et al., 2009; Hsu et al., 2014; Acharya and Xu, 2017).

3.2. The real economy effects of unconventional monetary policy

15. The contribution of Foley-Fisher et al. (2016) is closest to our paper in terms of

combining the risk-taking channel of unconventional monetary policy and its effects on

firm-level investment. Contrary to us, their analysis is confined to a specific part of the

unconventional monetary policy program of the Fed, namely the Maturity Extension

Program (MEP). Supported by evidence showing that the riskiness of newly issued

corporate bonds increased significantly during the MEP program, the authors confirm

that expansionary monetary policy was accompanied by an increase in risk-taking.

Moreover, the authors show that the MEP boosted investment for financially constrained

firms, although no connection is made between these results and their effects on

allocative efficiency and productivity. Another main distinction constitutes the fact that

Foley-Fisher et al. (2016) capture monetary policy through a dummy variable for when

the program occurred as opposed to our framework, which also captures differences in

the strength of the monetary policy stimulus.

16. As for the relation between unconventional monetary policy and financial

instruments, we compare to Lhuissier et al. (2018). Using a similar proxy SVAR

framework, the authors show that expansionary monetary policy incentivizes firms to

raise capital through corporate bonds rather than bank loans. Contrary to our study,

however, their empirical framework rests on aggregate data. Constrained by this

limitation, no implication is thus derived with regards to the effect of this change in

firms’ capital structures on allocative efficiency and firm-level productivity.

3.3. The economics of the risk-seeking channel

17. Expansionary unconventional monetary policies, by lowering interest rates

across the entire yield curve, may affect the risk perception and the risk-tolerance of

investors and induce them to bear a higher degree of risk in their portfolios. This risk-

taking behavior tends to be enhanced in an environment where not only current but also

future returns are very low due to unconventional monetary policy. Borio and Zhu

(2008) suggest that there are at least three ways through which the risk-taking channel

of monetary policy may operate: via credit expansion due to the increased valuation of

collateralized assets; via the decreased attitude towards risk due to the compression of

risk premia fostered by the central bank communication; and via the search for yield, as

narrowly defined, due to the incentives created by the structure of some economic

contracts in the financial industry.

│ 11

18. In this respect, the compensation scheme of professional money managers,

whose premium depends on whether the nominal returns of their portfolios are higher

than a predetermined specific threshold (Rajan, 2006), creates a motive to take on more

risk, particularly in a low interest environment. Similar incentives could be brought

about by remuneration scheme of pension funds or insurance companies, which have to

meet sticky nominal return targets in order to be able to meet future obligations. Bekaert

et al. (2013) provide evidence of a time-varying risk appetite by decomposing the VIX

index, which consists of a weighted average of implied volatilities of S&P

500 index options, into risk aversion and uncertainty. They show in a monthly VAR that

shocks to monetary policy significantly decrease risk aversion. Ceteris paribus, with

lower attitudes towards risk, investors take on more risk.

19. The literature provides ample empirical evidence for the risk-taking channel of

monetary policy. Jimenez et al. (2014) use data from the Spanish credit registry to show

that lower overnight interest rates lead banks to increase loan grants to ex ante riskier

borrowers. Ioannidou et al. (2014) exploit the fact that in the dollarized economy of

Bolivia monetary policy is exogenous to show that expansionary monetary policy

increases loans grants to firms with ex ante lower credit ratings. Similarly, Morais et al.

(2015) show that accommodative foreign monetary policy induces foreign banks to

grant more loans to Mexican firms with ex post higher default risk, thereby suggesting

that risk-taking also operates internationally. Moreover, Hau and Lai (2016) exploit

differences of local tightness of monetary policy in the Euro area due to differences in

local inflation and show that in countries with lower real rates, investors significantly

favor to invest in ex ante riskier equity mutual funds.

4. Data

20. The analysis focuses on the corporate bond market in the United States from

2008 to 2016 to study the effects of unconventional monetary policy on primary

corporate bond markets and the real economy. Our sample consists of listed non-

financial companies. It is constructed using two major datasets: Dealogic, which

provides data on bond deals, and Worldscope, which contains annual balance sheet data.

4.1. Bond issuance data

21. Data on corporate bond deals are sourced from Dealogic, which provides

information on gross issuance in primary corporate bond markets broken down by credit

rating. The database provides information on each deal such as maturity date, years to

maturity, yield to maturity, rating, coupon rate, use of proceeds, etc. Additionally, it

includes sufficient information on the bond issuer and its parent company such that we

can identify the set of non-financial companies domiciled in the US issuing a bond. As

the focus of our study lies in analyzing the effects of US unconventional monetary policy

on corporate bond markets, this step is crucial as it allows to narrow down the set of

firms which are most impacted. By considering the nationality, the detailed business

class and the sector of both the issuer and the parent companies, we are able to identify

the relevant productive unit.3 This is a fundamental step as, although the analysis focuses

on non-financial companies, some non-financial parent companies may issue bond via

3 Annex A provides a more detailed description of the selection criteria and the steps followed to

clean the database. The final sample includes corporate bond deals issued in US, according to

Dealogic deal nationality, by listed non-financial firms belonging to the sectors listed in Annex

A Table A2.

12 │

financial vehicles. Without considering this information, one could, for instance,

wrongfully include an issuer who is a US financial vehicle with a foreign non-financial

parent corporation, or exclude a US-based non-financial parent corporation that obtained

resources through the issuance of a foreign-based or US-based financial vehicle. A

closer look at the data reveals that in about 6% of the cases non-financial private

corporations rely on domestic and foreign financial vehicles for their bond issuance

operations.4

22. As we intend to examine the risk-seeking channel induced by unconventional

monetary policy, we separate all corporate bond issuance in two broad credit rating

groups, high-yield and investment-grade bonds. Relative to their investment grade

counterparts, high-yield bonds offer greater coupon payments in order to compensate

investors for the higher risk of default. Typically, companies issuing HY bonds are

highly leveraged or are experiencing financial difficulties, though young and small

companies may also fall into this category as investors require higher returns in order to

be compensated for more volatile future cash-flows or for the uncertainty over their

business model due to a lack of good past performance. Once this selection is made, we

aggregate firm-level bond issuance by month and broad credit ratings in order to perform

the proxy-SVAR at the monthly level.

4.2. Balance-sheet data

23. Balance-sheet information and multifactor productivity are based on

consolidated firm-level data from Thomson Reuters’ Worldscope – a database which

provides data on all publicly traded firms in 75 countries starting in 1985. The main

advantage of Worldscope lies in its potential to draw from a rich set of firm-level

controls and its reliability, as listed companies mandatorily have to report a variety of

firm characteristics in their balance sheets, many of which are also publicly disclosed in

the form of Securities Exchange Commission (SEC) filings. Against this background,

Worldscope has been used frequently to study firm-level investment (e.g. Kang and

Piao, 2015; Dlugosch and Kozluk, 2017).

24. A main disadvantage of Worldscope is that it is constrained to listed firms and

thus excludes by design firms not traded on stock exchanges. In the context of our

analysis, one could argue that the restriction to publicly listed firms may be problematic

as firms entering the bond market do not per se need to be listed on a stock exchange. In

reality, the overwhelming majority of corporations issuing bonds however are either

publicly listed or the subsidiary of a listed firm. Thus, our sample of non-financial

companies with a rating and public listing represents 94.10% of the average yearly total

bond issuance over the 2008-2016 period.

25. Since the focus is on the US, this disadvantage is further diminished, given that

the majority of US firms seek access to external funds through market-based financing

as opposed to bank-based financing and thus, the majority of firms are actually publicly

4 To name one example, the parent company Ford Motor Co which is based in US issues bonds

both through issuers based in US such as Ford Motor Credit Co LLC and issuers based outside

US as Ford Motor Credit Canada Ltd or Ford Credit Australia Ltd.

│ 13

traded firms (Cihak et al., 2013; Demirguc-Kunt and Maksimovic, 2002).5 In sum, most

US firms in Dealogic are thus contained in the Worldscope database.

26. Issues arising from the underrepresentation of certain firm types and industries

could nevertheless emerge as the changing nature of investment behaviour of firms with

greater access to capital due to bond issuance could spread to the whole universe of firms

in the US, including very young or small, or more generally non-listed firms. In the

absence of available balance-sheet data for those, however, this paper only captures

dynamics within the sample of listed firms.

4.3. Matching

27. While the core of the analysis does not require the matching of the two datasets

per se, we extend the discussion of our results to investigate the link between firm and

bond characteristics (see section 5.3 and 5.4). In particular, we assess firms’ propensities

to issue high-yield bonds based on their productivity performance, and investigate

whether the use of proceeds from bond issuance differs across firms with different

productivity profiles. Since Worldscope and Dealogic operate with different company

code identifiers, we use the bond’s ISIN to single out a common identifier at the firm-

level.6 In cases where this proves impossible, we manually merge the two dataset by

names. After cleaning and filtering both datasets, we are able to match 95% of

observations in Dealogic with Worldscope records. In all cases, this relates to the status

of the firm issuing the bond, which had not yet been publicly listed and therefore could

not be covered by Worldscope. Dealogic data are aggregated at the yearly level to pair

them with the firm’s balance sheet data.

4.4. Productivity measurement

28. To calculate firm-level multi-factor productivity we use a Solow residual

approach, where MFP is measured as the residual of a standard Cobb-Douglas

production function. Along the lines of Gal (2013), data on value added are externally

imputed as the sum of factor incomes going to employees, (wL, i.e. average labour cost

times the number of employees), and to capital owners (rK, i.e. the rate of return times

the capital stock). While the number of employees and information on the capital stock

(measured as the deflated book value of fixed tangible assets) can be sourced from

Worldscope, data on the cost of employees by firm are not available, and thus obtained

as industry-average from the OECD STAN database. All monetary variables are deflated

using 2-digit US-specific industry deflators constructed from OECD STAN and national

accounts. As usual, measuring productivity comes with a number of caveats, most

notably, the unobserved quality and utilisation of inputs, as well as unobserved

differences in market power (Gal, 2013).

5 A number of explanations have been offered as to why access to corporate bond market tends

to be linked with listings, including management’s prior experience with public securities; stricter

corporate governance requirements making the form less prone to debt-related moral hazard

concerns in the eyes of investors; and the absence of additional costs for producing financial

statements in accordance with regulatory rules and requirements (OECD, 2015a).

6 ISIN: International Securities Identification Number (https://www.isin.org/isin/). CUSIP:

Committee on Uniform Securities Identification Procedures

(https://www.cusip.com/cusip/about-cgs-identifiers.htm)

14 │

5. Empirical approach

5.1. The proxy SVAR framework

29. We use a vector autoregression (VAR) framework with macroeconomic and

financial variables to study the effect of monetary policy on issuance of corporate bonds

in primary markets. The main empirical challenge lies in identifying the monetary policy

shock. Standard practice relies on timing restrictions in order to obtain monetary policy

shocks which are orthogonal to other shocks in the VAR. However, in a VAR setup that

simultaneously employs macroeconomic and financial variables this procedure appears

less appropriate as it would imply that financial variables react to monetary policy only

with a lag which seems implausible (Gertler and Karadi, 2015). Therefore we make use

of recent advancements in the literature on the identification of monetary policy and use

a proxy SVAR approach.

30. In particular, we follow Mertens and Ravn (2013) and Gertler and Karadi (2015)

who show that identification of a shock within a VAR can be achieved through a proxy

variable, i.e. information external to the VAR. As long as the proxy is correlated with

the shock we seek to identify but orthogonal to other shocks we can use instrumental

variable regressions to uncover the impact of a monetary policy shock on any other

variable.

31. In order to illustrate the underlying mechanics of the proxy SVAR, consider the

following simple example of a reduced-form VAR representation with two variables and

one lag 7:

(𝑦1,𝑡

𝑦2,𝑡) = (

𝑏11 𝑏12

𝑏21 𝑏22) (

𝑦1,𝑡−1

𝑦2,𝑡−1) + (

𝑢1,𝑡

𝑢2,𝑡) (1)

where 𝑦1 can be any economic or financial variable, 𝑦2 is an indicator for monetary

policy, 𝑏𝑖𝑗 reduced-form coefficients and 𝑢1and 𝑢2 reduced-form shocks. We can write

the reduced-form shocks as a function of the structural shocks, 𝑒1and 𝑒2:

(𝑢1,𝑡

𝑢2,𝑡) = (

𝑠11 𝑠12

𝑠21 𝑠22) (

𝑒1,𝑡

𝑒2,𝑡) (2)

32. In order to identify the monetary policy shock, we need to find estimates of the

structural coefficients 𝑠12 and 𝑠22. The proxy SVAR approach provides us with these

estimates if we can find a further variable external to the VAR which fulfills two

conditions. First, the instrument needs to be correlated with the structural monetary

policy shock, 𝑒2, but simultaneously uncorrelated to the other shock 𝑒1. We can then use

instrumental variable regressions to obtain an estimates of the ratio of 𝑠22

𝑠12:

𝑢1,𝑡 =𝑠22

𝑠12𝑢2,�̂� + 𝑤𝑡 (3)

7 The generalisation to more than two variables and more than one lag is straightforward and

does not change any of the above conclusions, see Gertler and Karadi (2015).

│ 15

where 𝑤𝑡is a white noise error term. Intuitively, the first-stage of the instrumental

variable approach forms the prediction 𝑢2,�̂� which can be interpreted as movements in

monetary policy not anticipated through variation in other variables, i.e. surprises in

monetary policy. The coefficient 𝑠12 can be calculated algebraically from the reduced-

form variance-covariance matrix without further identifying restrictions. Eventually we

obtain an estimate of 𝑠22, the structural coefficient of interest.

33. Following the approach of Gertler and Karadi (2015) we use high-frequency

surprises in interest rates around Fed announcements as instruments to identify the

monetary policy shock. Since we do not have access to intra-day data, in line with

Hansson and Stein (2015), we calculate surprises in daily windows around Fed

announcements. The underlying identifying assumption is thus that no other factors

besides monetary policy affect interest rates within this window. As a last step, we need

to mensualize the high-frequency surprises around Fed announcements to match the

monthly frequency of the VAR. Here, we follow Gertler and Karadi (2015) and create

monthly average surprises by cumulating all FOMC day surprises, then take monthly

averages of this series and get monthly average surprises through a further first-

difference transformation. This procedure also takes into account that not all FOMC

meetings take place on the same day within a month.

34. Our identification strategy relies on the assumption that the external instruments

are uncorrelated with all other shocks than the monetary policy shock. However, two

potential confounding factors could potentially impede causal inference. First, some

important economic releases occur on the same day as the FOMC announcement and

therefore could be picked accidently up by our instrumental variable regressions.

Second, the Fed might have knowledge that market participants do not have, and, as a

consequence, FOMC announcements could also trigger reactions by investors unrelated

to the conduct of monetary policy. Using intra-daily data, e.g. minute-by-minute data,

in narrow windows around the FOMC announcements could potentially, alleviate these

concerns. However, in the lack of access to intra-daily data, following previous research,

we use daily changes in interest rates around FOMC announcements (e.g. Hansson and

Stein 2015; Lhuissier and Szczerbowicz, 2017) to identify shocks to monetary policy.

Further, since we focus on long-term interest rates, Hansson and Stein (2015) argue that

since market participants only react to changes in monetary policy with a lag, daily

changes are preferable.

35. As we seek to analyze the impact of expansionary monetary policy on primary

and secondary bond markets, we use a monthly VAR consisting of the following vector

of variables:

[log 𝐼𝑃𝑡, log 𝐶𝑃𝐼𝑡, 𝑖𝑡 ,𝐻𝑌𝑡

𝐼𝐺𝑡, 𝑟ℎ𝑦,𝑡 − 𝑟𝑖𝑔,𝑡] (4)

where IP stands for industrial production, CPI for inflation, 𝑖𝑡 for a monetary policy

indicator (a government bond rate), 𝐻𝑌𝑡

𝐼𝐺𝑡 for the ratio of total high-yield issuance of US

domestic non-financial corporate bonds over total investment-grade issuance of US

domestic non-financial corporate bonds and 𝑟ℎ𝑦,𝑡 − 𝑟𝑖𝑔,𝑡 for the spread between yields

on non-financial HY corporate bonds and non-financial IG corporate bonds. An increase

in the HY-IG spread indicates that prices for HY bonds decreased relative to IG bonds,

and vice versa.

16 │

36. Using the proxy SVAR, we test whether expansionary unconventional monetary

policy increases investors’ appetite for risk in corporate bond markets. A significant

increase in 𝐻𝑌𝑡

𝐼𝐺𝑡 would show that issuance of high-yield corporate bonds reacts stronger

to expansionary unconventional monetary policy than investment-grade corporate

bonds, indicating an increase in riskiness of the composition of new US domestic non-

financial corporate bonds. In order to check whether the risk-seeking behavior is actually

driven by a change in risk appetites of investors, we further look at the HY-IG spread.

An increase in relative high-yield bond issuance (𝐻𝑌𝑡

𝐼𝐺𝑡) and a simultaneous decrease in

the spread (𝑟ℎ𝑦,𝑡 − 𝑟𝑖𝑔,𝑡) would provide evidence for greater investors’ demand of HY

corporate bonds, as the increased demand for HY bonds relative to IG bonds would drive

up the prices for HY bonds and decrease their yields.

37. Last, we experiment with a slightly different VAR specification where we

substitute 𝑟ℎ𝑦,𝑡 − 𝑟𝑖𝑔,𝑡 with the excess bond premium spread instead, 𝐸𝐵𝑃𝑡 , as in Gertler

and Karadi (2015). The Gilchrist and Zakrajasek (2012) excess bond premium spread is

an indicator for refinancing conditions on secondary corporate bond markets cleaned for

bond defaults and as such a proxy for refinancing conditions absent financial frictions

induced by the fear of default. By contrasting results from a VAR specification with

yield spreads and with the excess bond premium, we can analyse the mechanism through

which monetary policy induces risk-taking, i.e. through a reduction of yields or through

lower (perceived) default risk.

5.2. VAR Estimation and the choice of the monetary policy indicator and

instrument

38. We derive the structural coefficients of a monetary policy shock in two steps.

First, we estimate the reduced-form VAR equation by equation with OLS and save the

innovations from the reduced-form regressions. For each variable in the VAR but the

monetary policy indicator, we run an instrumental variable regression of the innovation

of the variable we want to examine, e.g. high-yield bond issuance relative to investment-

grade bond issuance, on the innovation of the monetary policy indicator. In this

instrumental variable regression, the first stage isolates movements in the monetary

policy indicator solely due to surprises in interest rate futures and thus disentangles the

surprise component from other confounding factors. Second, the estimated coefficients

are used to obtain the structural coefficient estimates that are used to study the impulse

response functions. These two steps allow to compute a time series of the part of 𝐻𝑌𝑡

𝐼𝐺𝑡

induced by UMP, which can be included in the firm-level reallocation model.

39. Since our focus is on the period of unconventional monetary policy in the US,

we seek to run the instrumental variable regressions over the same period. In line with

Lhuissier and Szczerbowicz (2017), we choose a starting point for our structural analysis

shortly before the announcement of the Fed of the first QE program, i.e. 2008M10. Since

we have longer time series for all variables in the VAR, we experiment with two

estimation windows for the estimation of the reduced-from coefficients. First, we

estimate the reduced-form on the same time period as in the structural analysis, i.e.

│ 17

starting in 2008M10.8 Second, as a robustness check, we use the full time series available

in Dealogic from 1993M1 to estimate the reduced-form.

40. In addition to the dates of FOMC announcements, we extend the set of important

dates with a list of further Fed communications from Rogers et al. (2014). This list

includes speeches by B. Bernanke with further information on the future conduct of

monetary policy. The additional dates added to the list of FOMC announcement dates

are 25/11/2008 (Fed announces purchases of MBS and Agency bonds), 01/12/2008

(Bernanke states Treasuries may be purchased), 27/08/2010 (Bernanke Speech at

Jackson Hole), 15/10/2010 (Bernanke Speech at Boston Fed), 26/08/2011 (Bernanke

Speech at Jackson Hole), 31/08/2012 (Bernanke Speech at Jackson Hole) and

22/05/2013 (Bernanke Testimony). However, results also hold if we do not add these

further dates.

41. We use a generic four-lag VAR to determine the optimal combination of

monetary policy indicator and instrument. As we focus on unconventional monetary

policy in the US and consistent with the view that QE should flatten the yield curve, we

restrict the set of potential monetary policy indicators to long-term Treasury bonds. In

order to avoid running into a problem with weak instrument, we estimate a reduced-

form VAR for different indicators of monetary policy, i.e. for Treasury bonds with

different long maturities, and then check the statistical properties of the results of the

first-stage of the instrumental variable regression, i.e. the regression of the innovation

in the monetary policy indicator on the instrument. Since we do not have access to

intraday data for US long-term government bonds, we follow Hansson and Stein (2015)

and compute daily surprises9 around Fed announcements.

5.3. Monetary policy shocks and capital reallocation

42. In the second step, we use our estimate of the evolution in the 𝐻𝑌𝑡

𝐼𝐺𝑡 ratio due to

monetary policy to study allocative efficiency on capital markets, i.e. whether more

productive firms increase their investment by larger amounts as compared to less

productive firms. We obtain a time series of the 𝐻𝑌𝑡

𝐼𝐺𝑡 ratio that is solely due to surprises

in monetary policy by:

𝐻𝑌𝑡𝑀𝑃

𝐼𝐺𝑡𝑀𝑃 = 𝑠

𝐻𝑌𝐼𝐺 × 𝑢𝑡

𝑀�̂� + 𝑤𝑡 (5)

where 𝐻𝑌𝑡

𝑀𝑃

𝐼𝐺𝑡𝑀𝑃 is the new time series of

𝐻𝑌𝑡

𝐼𝐺𝑡 ratio due to monetary policy, 𝑢𝑡

𝑀�̂�the variation

in the reduced-form innovation of the monetary policy indicator which is solely driven

by the surprises and thus orthogonal to the other shocks in the VAR, and 𝑠𝐻𝑌

𝐼𝐺 the

8 To make sure the lags in the VAR do not move the estimation window beyond the 2008M10

starting point and thus potentially excluding relevant FOMC announcements, we choose a

starting point for the reduced-form estimation such that incorporating all lags, the first data point

of the innovations corresponds to 2008M10.

9 Specifically, we compute the surprises around a Fed announcement as 𝑠𝑢𝑟𝑝𝑟𝑖𝑠𝑒𝑡 = 𝑟𝑡+1 − 𝑟𝑡−1,

where 𝑟 is the yield on a government bond.

18 │

structural coefficient we obtained from the instrumental variable regressions and from

information of the reduced-from variance-covariance matrix. Since the firm-level model

relies on balance-sheet data which usually comes at an annual frequency, we aggregate

the estimated monthly time series of 𝐻𝑌𝑡

𝑀𝑃

𝐼𝐺𝑡𝑀𝑃 by summing all values within a year.10 For

matters of simplicity, we will refer to the ratio 𝐻𝑌𝑡

𝑀𝑃

𝐼𝐺𝑡𝑀𝑃 as “HY shock” for the rest of the

analysis.

43. To test for potential distortionary effects of UMP-driven changes in the HY

shock on resource allocation, we augment the canonical model of firm dynamics (see

Foster et al., 2016; Decker et al., 2016) with the HY shock, obtained from the proxy

SVAR. The basic idea behind these models is that conditional on firm size, firms with a

higher productivity should attract more (financial) resources and grow faster. As a result,

highly productive firms would tend to grow at the expense of unproductive firms. In

equilibrium more productive firms should be characterised by higher market shares, a

concept known as allocative efficiency (Andrews and Cingano, 2014). An economy

where allocative efficiency is distorted would imply that resources get trapped in

unproductive firms therefore impeding the growth of productive firms and putting

downward pressure on aggregate productivity growth. If monetary policy induced HY

bond issuance had an influence on the capital reallocation dynamics, it would thus be

captured by the interaction between the HY shock and lagged multi-factor productivity.

44. We embed our empirical framework in an otherwise standard Tobin’s Q

investment model. The Tobin Q model implies that firms should invest if the market

value of a marginal unit of capital exceeds its replacement costs. Following the literature,

Tobin’s Q is calculated as the sum of market value of assets and debt divided by total

assets. All information regarding potential investment projects is summarized in Q, as

the Q quantifies the increase in the discounted firms’ profits and thus the increase in

total firm value if a firm buys an additional unit of capital. However, the Q model relies

on certain assumptions and may not well control for firm-level investment demand when

asset prices are subdued due to deviations from fundamental value, i.e. in times of fire

sales. Further, firms might have incentives to invest aside internal investment demand,

e.g. through taxes, which are not captured by the Q model.

45. The resulting baseline specification is as follows:

𝐼𝑖,𝑗,𝑡

𝐾𝑖,𝑗,𝑡−1

= 𝛼 + 𝛽0 × 𝑄𝑖𝑗,𝑡−1 + 𝛽1 × 𝑀𝐹𝑃𝑖𝑗,𝑡−1 + 𝛽2 × 𝑀𝐹𝑃𝑖,𝑗,𝑡−1

× 𝐻𝑌𝑠ℎ𝑜𝑐𝑘𝑡 + ∑ 𝛾𝑙𝑥𝑖,𝑗,𝑡

𝑙

+ 𝛿𝑗 + 𝛿𝑡 + 𝜖𝑖𝑡 (6)

where 𝐼𝑖,𝑗,𝑡

𝐾𝑖,𝑗,𝑡−1 denotes tangible investment of firm i in sector j at time t scaled by the

tangible capital stock in period t-1, 𝑀𝐹𝑃𝑖 is the index of multi-factor productivity of

firm i, and 𝑥𝑖,𝑗,𝑡 stands for other firm-level characteristics. Our baseline specification

includes Tobin’s Q and controls for firm-specific demand-effects, measured by

contemporaneous sales over lagged capital (see IMF, 2015 for comparison). To diminish

10 Results are qualitatively similar if we aggregate the monthly values by taking their yearly

mean.

│ 19

concerns of reverse simultaneity, we use lagged values of 𝑀𝐹𝑃 and Tobin’s Q. To check

for robustness, we also augment the baseline specification with a broader set of control

variables in order to alleviate concerns with omitted variable bias.

46. Allocative efficiency would imply that 𝛽1, the sign of the estimated coefficient

of 𝑀𝐹𝑃𝑖𝑗,𝑡−1, is positive, so as to reflect that more productive firms attract more

resources. The expected sign of the interaction between productivity and additional HY

corporate bond issuance due to monetary policy, 𝛽2, on the other hand, is a priori

ambiguous, given that the relationship between productivity and credit ratings tends to

follow an inverse U-shape (see Figure 2). The main point of our analysis it to establish

whether low, or high productive firms effectively benefit from an increased demand in

HY bonds relative to IG bonds and as such shed light on the impact of unconventional

monetary policy on allocative efficiency. A positive and significant 𝛽2 coefficient would

suggest that search for yield induced by monetary policy benefits mostly highly

productive firms, thereby pointing to an increase in allocative efficiency. The reverse

would be suggested by a negative and significant coefficient, though with caveats that

will be discussed later.

47. In the baseline specification, we control for unobserved shocks common to all

firms and sector-specific shocks through the use of year and sector fixed effects (𝛿𝑗 and

𝛿𝑡 ). These could emanate from sector-specific regulations affecting the overall

attractiveness to invest in one particular sector, or from cyclical influences, pertaining

to the whole economy at a specific point in time. We also experiment with a more lenient

approach that allows us to further incorporate the HY shock as a standalone variable, by

using sector fixed effects only. To the extent that patterns of resource allocation tend to

reflect factor adjustments within sectors (see Foster et al., 1996), the most challenging

fixed effects structure we can employ in this analysis includes industry, year and

industry-year fixed effects.

48. We assess the sensitivity of our results in a number of ways. First, we try to

reduce the possibility that the coefficient estimates could suffer from omitted variable

bias. In particular, we augment our regression baseline to account for a firm’s age and

size, as measured by number of employees. Similarly, omitted variables bias could arise

due to the possible influence of confounding macroeconomic factors. Our baseline

specification at most captures shifts at the industry, year or industry-year level.

Macroeconomic shifts pertaining to the whole economy, however, are not taken into

consideration, and could bias the effect of UMP (captured by HY shocks) on capital

accumulation. To address this concern, we further add an interaction between GDP (as

a catch-all variable) and MFP to our regressions.

49. Expansionary unconventional monetary policy might not only affect the amount

of bond issuance but could also affect the likelihood to actually issue a corporate bond.

Therefore our model should account for the selection bias resulting from firms who start

to issue bonds due to expansionary monetary policy. Thus, we use a two-stage Heckman

regression model (Heckman, 1979), in which the first stage assesses firms’ propensity

to issue high-yield bonds and use this information in a second stage regression, which is

equivalent to the baseline specification to correct for the selection bias.

20 │

6. Results

6.1. Did expansionary monetary policy increase demand-driven risk-seeking

behaviour in corporate bond markets?

50. Table 1 shows the results of the first-stage regressions for two different monetary

policy indicators, the 5Y and 10Y real rate. Column 1 of Panel A and B depict that both

mensualized daily surprises in 3Y real rates and 5Y real rates tend to be highly correlated

with a 5Y real rate monetary policy indicator. Further, we find similar results for the

10Y real rate in column 2 of Panel A and B. The following columns then show that these

results are robust to changes in the VAR specification and different estimations

windows. Specifically, column 3 and 4 of Panel A and B include a different measure of

the spread between high-yield and investment-grade bonds. Lastly, column 5 and 6 show

that the results change little when we estimate the reduced-form VAR over the full

period.

51. In order to choose the optimal combination we opt for a decision criterion that

relies on the F-test of the joint model. Stock, Wright and Yogo (2002) recommend a

threshold of 10 for the F-statistic of the first-stage regression such that problems with

weak instruments do not impede the credibility of our results. Then, in line with Gertler

and Karadi (2015) and Lhuissier and Szczerbowicz (2017), we choose the combination

of monetary policy indicator and instrument with the highest F-statistic for our preferred

VAR specification, i.e. the model from column 1 in Panel B. In the following, we will

thus use the 5Y real rate as a monetary policy indicator and instrument it with the

changes in 3Y real rates.

│ 21

Table 1. Structural identification: First-stage results

Panel A: T5Y Real

T5Y Real T10Y Real T5Y Real T10Y Real T5Y Real T10Y Real

(1) (2) (3) (4) (5) (6)

IV: T5Y Real 0.685*** 0.599*** 0.678*** 0.571*** 0.737*** 0.662***

(0.025) (0.023) (0.031) (0.025) (0.027) (0.024)

Observations 99 99 96 96 99 99

R-squared 0.213 0.205 0.191 0.181 0.233 0.238

F-test model 19.06 15.56 14.86 13 20.19 18.54

Rf. VAR estimation period 08-16 08-16 08-16 08-16 93-16 93-16

(HY-IG) yield spread Y Y N N Y Y

GZ Excess bond spread N N Y Y N N

Panel B: T3Y Real

T5Y Real T10Y Real T5Y Real T10Y Real T5Y Real T10Y Real

(1) (2) (3) (4) (5) (6)

IV: T3Y Real 0.661*** 0.537*** 0.678*** 0.525*** 0.704*** 0.583***

(0.023) (0.021) (0.026) (0.021) (0.026) (0.024)

Observations 99 99 96 96 99 99

R-squared 0.201 0.166 0.194 0.155 0.215 0.187

F-test model 19.09 13.42 17.82 13.22 18.95 14.31

Rf. VAR estimation period 08-16 08-16 08-16 08-16 93-16 93-16

(HY-IG) yield spread Y Y N N Y Y

GZ Excess bond spread N N Y Y N N

Note: This table presents first-stage results of the instrumental variable regressions of reduced form VAR

innovations on reduced form VAR innovations in different monetary policy indicators. We use mensualized

daily surprises of interest rates around FOMC announcements and some important Fed speeches as an

instrument for reduced form VAR innovations in different monetary policy indicators. The VAR consists of

five variables, an indicator for monetary policy, industrial production, CPI, a corporate bond spread indicator

and the ratio of new HY issuance over new IG issuance and is estimated with 4 lags. We use monthly data

from October 2008 to December 2016, and up to August 2016 for VAR specifications with GZ’s excess bond

spread which covers a shorter time span. Robust standard errors in parentheses. ***/**/* denotes significance

at the 1%/5%/10% level.

52. Figure 4 shows the resulting time series of the HY shock and the yearly averages

of monetary policy surprises. The negative correlation implies that as positive surprises

indicate a more contractionary stance of monetary policy relative to expectations, we

would expect HY shock to fall as investors have less incentive to take on risk and thus

likely have a higher demand for IG relative to HY bonds leading to the fall in their ratio.

Further, the time series of the HY shock shows that in the beginning of the period of

unconventional monetary policy, as monetary policy gets increasingly expansionary,

HY bond issuance was more pronounced relative to IG bonds.

22 │

Figure 4. The ratio of HY/IG due to monetary policy shock

Note: This graph shows the HY/IG ratio due to monetary policy shocks and the shocks as estimated by the

proxy SVAR. We construct the HY/IG ratio due to monetary policy shocks by multiplying the estimated

second-stage coefficient of the instrumental variable regression with the monetary policy indicator predicted

by the surprises. The monetary policy indicator is the 5Y real rate, its instrument the mensualized daily

surprise around FOMC announcements and some speeches of the 3Y real rate. The graph on surprises in

3Y real rates shows average monthly surprises within a year.

53. We then proceed to estimating the structural coefficients for all variables in the

VAR and analyse their propagation using impulse-response functions. Figure 5 shows

impulse responses for two different VAR specifications. Both models include the 5Y

real rate as indicator of unconventional monetary policy, CPI and industrial production,

the HY shock and a spread measure. Whereas model 1 includes the spread in yields

between HY and IG bonds, model 2 uses the excess bond premium index. All panels in

Figure 5 depict the impulse responses of an endogenous variable to a negative surprise

in 5Y real rates which we interpret as an expansionary monetary policy shock. For each

graph, the solid black line shows the median response, the dashed lines the 90%

confidence bands. We compute the impulse responses using a wild bootstrap with 2,000

repetitions (Gertler and Karadi (2015).

54. Panels A.1 and A.2 of Figure 5 show the response of the 𝐻𝑌𝑡

𝐼𝐺𝑡 ratio due to an

expansionary monetary policy corresponding to a fall in 5Y real rates of around 18 basis

points. The 𝐻𝑌𝑡

𝐼𝐺𝑡 ratio increases by around 10 percentage points after an expansionary

monetary policy shock, indicating a higher degree of riskiness of new corporate bond

issuances. In the second month after the expansionary monetary policy shock, the ratio

actually decreases potentially pointing to a lagged reaction of IG but not HY bonds.

Effects in the first two periods for both model specifications are significant at the 95%

level. However, the cumulated effect after the first two periods shows a significant

increase in the 𝐻𝑌𝑡

𝐼𝐺𝑡 ratio of around 5 percentage points. With the median value of the

𝐻𝑌𝑡

𝐼𝐺𝑡

ratio at approximately 60%, this increase corresponds to an increase in the ratio of

around 8%. After the first two periods, the responses revert back to zero within 7-12

months.

│ 23

55. Turning to the spread between HY and IG bond yields, Panel B.1 of Figure 5

shows a decrease in the spread of around 13 basis points. This decrease implies that

prices of HY bonds have risen stronger than IG bonds putting downward pressures on

the yield spread. This points towards an increased risk appetite of investors. The

response of the spread slowly converges back to 0 over the next three years. The

combination of increasing quantities and decreasing yields in the high-yield bond market

relative to the investment grade segment supports the risk taking hypothesis due to

unconventional monetary policy expansion.

56. Panel B.2 of Figure 5 shows the VAR specification with the excess bond

premium index as used in Gertler and Karadi (2015). After an expansionary monetary

policy shock, this index falls by a small amount; however the response is significant at

the 90% level only after three quarters.11 What is relevant in our set-up is that the relative

high-bond issuance is stable considering the more generalized index of refinancing costs

absent of default risk. Further, the fact that the spread between yields on high-yield and

investment-grade corporate bonds decreases significantly but not the default free excess

bond premium spread suggests that unconventional monetary policy affects primary

corporate bond markets through a reduction in perceived default risk rather than a

decrease in refinancing conditions.

57. Figure B.1 in the appendix shows impulse-response functions for the other

variables in the VAR. Panel B shows the response for CPI of the VAR specification

including the spread between HY and IG yields to an expansionary monetary policy

shock. Consistent with standard economic theory, prices rise significantly by around

0.9% and remain rather elevated through the whole 36 months considered for the

impulse response function. Output slightly decreases in the first period but then steadily

rises up to month 12 then converging back to zero. However, similar to Gertler and

Karadi (2015), the responses of industrial production are not significant at the 90% level.

58. In conclusion, we find qualitatively similar results for both VAR specifications

considered. Importantly the 𝐻𝑌𝑡

𝐼𝐺𝑡 ratio shows a significant increase in the first period.

Further, if we estimate the reduced-form VAR for the full time period for which we have

data, i.e. for 1993M01-2016M12 instead of 2008M06-2016M12, we obtain similar

results (Annex Figure B.2.).12

59. We find qualitatively similar results when we compare our analysis to Lhuissier

and Szczerbowics (2017) who also study the effect of expansionary unconventional

monetary policy on US corporate bond issuance in a proxy SVAR framework. However,

they focus more on capital structure decisions, i.e. on the choice between corporate bond

issuance and bank loans, but do not break down corporate bond issuance by credit rating

and as thus remain silent on risk-taking of investors on corporate bond markets. They

also find that expansionary monetary policy increased bond issuance, though they only

measure total issuance. They find significant responses for output and the excess bond

11 Gertler and Karadi (2015) find a negative and statistically significant effect on the EBP already

at impact, however, a direct comparison is no straightforward as they focused conventional

monetary policy using a 1Y nominal government bond with estimations done for a longer time

period.

12 In particular, the 𝐻𝑌𝑡

𝐼𝐺𝑡 ratio and the spread between yields between HY and IG bonds are

statistically significant and with the expected sign further underlining the robustness of our

results.

24 │

premium index with 68% confidence bands which do not contradict our findings of no

statistical significance based on more conservative confidence intervals.

Figure 5. Risk-seeking in corporate bond markets after an unconventional expansionary

monetary policy shock

Panel A.1. Model 1: HY/IG Ratio Panel A.2 Model 2: HY/IG Ratio

Panel B.1. Model 1: HY-IG Spread

Panel B.2. Model 2: GZ’s excess bond premium

Note: This figure depicts the impulse response functions of a one-unit standard deviation expansionary surprise in

monetary policy. We identify the monetary policy shock using an instrumental variable approach. The IRF are shown

for VAR specifications with the 5Y real rates as policy indicator and mensualized daily surprises in 3Y real rates as

instruments. Both VAR models include the logarithm of industrial production and CPI as further variables. The reduced-

form VAR is estimated over the 2008M06-2016M12 period. We obtain the IRFs using a wild bootstrap with 2,000

repetitions. The solid black line shows median responses. The dashed lines the lower and upper bounds of a 90%

confidence band.

6.2. Did monetary policy shocks distort the capital reallocation process?

60. Table 2 displays the baseline estimates of equation 6 in which we analyse the

effects of unconventional monetary policy shocks on capital reallocation by allowing

for changes to the HY shocks to directly affect firm-level investment. As the HY shock

relies on aggregate data, the inclusion of year fixed effects comes at a cost of dropping

the HY shock (column 2). The main variable of interest the interaction between lagged

MFP and HY shock. Column 3 further incorporates sector-year fixed-effects controlling

for sector-level confounding variables. To control for potential error correlation within

firms, robust standard errors are clustered in this dimension in all specifications.

-0.2

-0.1

0

0.1

0.2

0.3

1 4 7 10 13 16 19 22 25 28 31 34

Pe

rce

nta

ge p

oin

ts

Months

-0.2

-0.1

0

0.1

0.2

0.3

1 4 7 10 13 16 19 22 25 28 31 34P

erc

en

tage

po

ints

Months

-0.4

-0.3

-0.2

-0.1

0

0.1

0.2

1 4 7 10 13 16 19 22 25 28 31 34

Pe

rce

nta

ge p

oin

ts

Months

-0.4

-0.3

-0.2

-0.1

0

0.1

0.2

1 4 7 10 13 16 19 22 25 28 31 34

Pe

rce

nta

ge p

oin

ts

Months

│ 25

61. In line with the hypothesis that more productive firms should grow faster than

less productive firms, a higher productivity is positively associated with firm-level

investment across all specifications, suggesting that more productive firms are able to

attract relatively more capital. Put differently, capital reallocation within the same

industry overall is productivity-enhancing. Moreover, Tobin’s Q and firm-level

demand-effects also display the expected positive and significant association,

suggesting that the framework model accurately captures both firm-level investment

decisions and capital reallocation dynamics.13

Table 2. Reallocation - Baseline results

Dependent variable: capital expenditures

1 2 3

L.MFP 0.1117*** 0.1107*** 0.1160***

(0.0102) (0.0103) (0.0104)

L.MFP#HY shock -0.4641*** -0.4654*** -1.2294***

(0.0915) (0.0923) (0.2763)

HY shock -1.5932***

(0.1383)

L.Q1 0.0067*** 0.0069*** 0.0066***

(0.0021) (0.0021) (0.0021)

Sales/Capital 0.0042*** 0.0042*** 0.0041***

(0.0004) (0.0004) (0.0004)

Observations 21,632 21,632 21,626

R-squared 0.1458 0.1481 0.1580

Adj. R2 0.144 0.146 0.146

Industry FE Yes Yes Yes

Year FE No Yes Yes

Industry-Year FE No No Yes

Level of clustering Firm Firm Firm

Note: This table reports the baseline results where capital expenditures over lagged fixed assets are regressed

on lagged values of firm-level multi-factor productivity (Panel A) or the marginal revenue product of capital

(Panel B), their respective interaction with the HY shock obtained from the first part of the analysis, the

firm's lagged market value (measured using Tobin's Q) and a measure of aggregate demand. Industry refers

to 2-digit level detail according to NACE Rev. 2, covering the non-farm non-financial business sector

(industry codes 10-83, excluding 64-66), year to the period 2008-13. Robust standard errors are reported in

parenthesis. *** denotes statistical significance at the 1% level, ** significance at the 5% level, and * and

the 10% level.

Source: Calculations based on Thomson Reuters Worldscope and Dealogic.

62. The negative and significant coefficient of the interaction between MFP and the

HY shock indicates that expansionary unconventional monetary-policy may have

diminished allocative efficiency of the economy by increasing risk-seeking in the bond

market, and this is robust to different specification of the fixed effects structure. While

high productive firms thus generally seem to invest more than their low productive

peers, this effect is mitigated by the changes in the composition of the riskiness of

13 Corporate bond issuance at the sector-level could deviate from aggregate trends. Thus, the

results from the reallocation model might solely come from specific sectors and not necessarily

generalize to the whole economy. In order to show the robustness of our results, we drop

iteratively the five sectors with the largest increases in bond issuance over the 2008-2016 period

and re-estimate all reallocation models. Results hold robust.

26 │

corporate bonds. The changes to the HY shock itself are associated with an overall

decline in investment (column 1), though the total effect of changes in the HY shock

must be considered in relation with its interaction term (see below). Overall, estimation

results suggest that the changes in the composition of riskiness of corporate bond

issuance due to the risk-seeking channel of monetary policy transmission may constrain

the growth of more productive firms, in turn reducing aggregate MFP via lower

allocative efficiency.

63. To get a sense of the overall direction and magnitude implied by the coefficient

estimates, Figure 6 displays the findings graphically. Panel A represent the marginal

effects obtained from column I (Table 2), whereas Panel B is based on estimates of

column III (Table 2), in which the standalone effect of HY shocks is absorbed by year

fixed effects. If interpreted causally, the results suggest that a one standard deviation

increase of the HY shock would lead to a 2 percentage point increase in capital

expenditures for firms at the 25th percentile of the MFP distribution, and a 2 percentage

point decrease for firms at the 75th percentile. Low productive firms therefore increase

their investment following a HY shock, while high productive firms reduce their

investment.

Figure 6. Marginal effects of increased bond issuance due to monetary policy surprises on

firm-level capital expenditures

Panel A: Total effect Panel B: Differential effect

Note: This graph shows the marginal effects of a one standard deviation increase in the HY shocks that were

obtained in the first part of this paper for different centiles of the MFP distribution. Panel A reflects the

marginal effect calculated based on Table 2, column 1, while Panel B represents the marginal effects

calculated from Table 2, column 3 (sector, year, and sector-year fixed effects).

Source: Calculations based on Thomson Reuters Worldscope and Dealogic.

Crowding out effects

64. One explanation for the observed differences in investment following a shift in

the composition of riskiness on corporate bond markets could be the presence of

crowding-out effects. These effects could occur if the share of HY bonds issued in a

reaction to expansionary monetary policy was mostly associated with low productive

firms that rely more on corporate bond markets to fund new projects. As a result, these

firms would be able to acquire the financial means to increase capital spending on goods

and services that were previously out of reach. In turn, the increased demand for these

goods and services would drive up their price. All other firms would thus face higher

-.2

-.1

0.1

.2e

ffe

ct o

n inve

stm

ent

0 2 4 6 8 10MFP distribution

estimated marginal effect 95% confidence interval

95% confidence interval

Marginal effect of MP shock on investment: 2008-16

-.4

-.2

0.2

.4.6

effe

ct o

n inve

stm

ent

0 2 4 6 8 10MFP distribution

estimated marginal effect 95% confidence interval

95% confidence interval

Marginal effect of MP shock on investment: 2008-16

│ 27

procurement costs, possibly forcing them to refrain from their purchase. In this case, one

would observe a relative decline in investment for firms that are not issuing bonds.

65. Without firm-level data on procurement costs, we cannot directly test this

hypothesis, however we can check the underlying premise, i.e. whether low productive

firms have a higher likelihood to issue bonds. To that end, we use probit models to

analyse whether firms’ propensity to issue bonds differs by the level of MFP. If low

productive firms on average tend to issue more bonds, we can provide tentative

empirical evidence pointing towards a crowding-out mechanism. We test this conjecture

using a reduced-form probit regression, where the variable of outcome is defined as a

dummy capturing if a firm has issued any bond in year t, and whether this bond was

categorized as high-yield or investment-grade. All other control variables are identical

to the set of control variables used in our baseline estimation.

66. In line with our conjecturing, Table 3 shows that high MFP firms have a

significantly lower propensity to issue corporate bonds in general, but that these effects

are particularly pronounced for HY bonds (see column 2). Moreover, the non-linear

marginal effects obtained from this probit regression displayed in Figure 7 suggest that

a one percent increase in MFP significantly reduces the probability of issuing a HY

bond, especially at low levels of MFP. We interpret these results as evidence that the

subsample of firms which gained access to capital markets due to investors’ increasing

risk-seeking behaviour were indeed mainly low productive firms, supporting the idea

that a crowding-out effect may have taken place.

Figure 7. Marginal effects of increasing MFP on the propensity to issue HY bonds

Note: This graph shows the marginal effect of increasing MFP by one percent on the likelihood to issue HY

bonds for different starting points contained in our sample. Results are based on the probit regression

displayed in Table 3. The red point represents the point estimate, while the ranges report 95% confidence

intervals.

Source: Calculations based on Thomson Reuters Worldscope and Dealogic

28 │

Table 3. Crowding-out effects

The effect of firm-level characteristics on the propensity to issue bonds

Issuance dummy All bonds HY bonds IG bonds

L.MFP -0.191*** -0.198*** -0.142*** (0.0348) (0.0434) (0.0405)

L.Q1 -0.0285*** -0.0178** -0.0484*** (0.00769) (0.00811) (0.0121)

Sales/Capital -0.00158 -0.000206 -0.00323** (0.00107) (0.00130) (0.00162)

Observations 21,642 21,642 21,414

Issuance dummy All bonds HY bonds IG bonds

L.MFP -0.191*** -0.198*** -0.142***

(0.0348) (0.0434) (0.0405)

L.Q1 -0.0285*** -0.0178** -0.0484***

(0.00769) (0.00811) (0.0121)

Sales/Capital -0.00158 -0.000206 -0.00323**

(0.00107) (0.00130) (0.00162)

Observations 21,642 21,642 21,414

Note: Estimates from probit regression. Values are marginal effects. The coefficient corresponds to the

impact of a change in the explanatory variable on the probability to issue bonds at the mean of the

independent variable. Industry refers to 2-digit level detail according to NACE Rev. 2, covering the non-

farm non-financial business sector (industry codes 10-83, excluding 64-66); year to the period 2008-13. All

regressions include year and industry fixed effects. Robust standard errors are reported in parenthesis. ***

denotes statistical significance at the 1% level, ** significance at the 5% level, and * and the 10% level.

Source: Calculations based on Thomson Reuters Worldscope and Dealogic.

6.3. Robustness checks

67. To diminish endogeneity concerns, we perform several robustness tests. First,

we account for a potential selection bias, which could affect our baseline estimates if the

propensity to issue bonds was correlated with the effect of changes in the HY shock on

investment. For instance, changes in the HY shock may affect not only the amount of

investment of already issuing firms, but also the propensity of other firms to issue bonds

in order to exploit the higher appetite for risk of investors.

68. To address this possibility, we run a two-step Heckman selection procedure to

our baseline model. This consists in estimating first a firm’s propensity to issue high-

yield bonds based on the same set of explanatory variables as in our baseline

specification, and subsequently using the obtained correction factor (the so-called Mill’s

ratio) in the second stage estimation of the baseline specification. We find the first stage

results to be credible (see right column of Table 4)14 corroborating the results obtained

from the probit model in section 5.2, and second stage results to hold robust to adding

the correction factor. Therefore, selection bias is not likely to be a serious issue

influencing the interpretation of our results.

14 Credible estimates of a Heckman selection model require at least one variable to appear with

a non-zero coefficient in the selection equation and not in the equation of interest (Caldera-

Sànchez and Andrews, 2011).

│ 29

69. Next, we address concerns related to omitted variable by augmenting the

baseline equation with additional firm-level characteristics, notably firm age and size

(Table 5, Panel A). The existing literature treats both as potential determinants of firm-

level investment, not least since they are considered reliable predictors of firm-level

credit constraints (Gala and Julio, 2016). The inclusion of these controls comes at the

cost of losing significance for Tobin’s Q, which is strongly correlated with firm size.

However, the interaction term between HY shocks and MFP remains unaffected.

70. Endogeneity bias could also arise due to the confounding macroeconomic

factors, such as if the identification of the first stage HY shock was not purely

exogenous, hence capturing broader cyclical effects. The fixed effect structure controls

for omitted variable bias that could induce shifts at the industry or year level, but does

not control for omitted factors at the more aggregate level. To account for this, we

include an additional interaction term between the output gap (a catch-all variable) and

MFP. Except for the most demanding fixed-effect structure, the results displayed in

Table 5 (Panel B), however, are robust to this change.

71. Lastly, the interaction of MFP and the HY shock could pick up confounding

firm-level factors. For instance, changes to the HY shock could also change the

sensitivity of firm-level investment to underlying firm-level fundamentals like Tobin’s

Q or sales. Including interactions of the HY shock with Tobin’s Q and firms’ sales is a

way to test the robustness of the MFP interaction to these possible firm-level effects.

Table 5 (Panel C) shows that the interaction of the HY shock and MFP remains robust

to this extended specification.15

15 The interaction of Tobin’s Q is significant and negative implying that changes in the corporate

bond market risk reduce the sensitivity of investment to Tobin’s Q. The interaction with firm-

level sales is not significant.

30 │

Table 4. Robustness I: A Heckman selection model

Second stage First stage

Dependent variable: Capital expenditures

Dependent variable: Issuing HY bonds

L.MFP 0.115*** L.MFP -0.155***

(0.011) (0.036)

L.MFP#HY shock -0.385*** L.MFP#HY shock 0.603***

(0.083) 0.232

L.Q1 0.008*** L.Q1 -0.032***

(0.002) (0.011)

Sales/Capital 0.004*** Sales/Capital -0.003**

(0.000) (0.001)

Dummy HY Issuance 0.811***

(0.058)

Observations 18789 Observations 18789

Note: This table presents first and second stage results of a Heckman selection model, where the first stage

measures firms’ propensity to issue HY bonds and the second stage models firm-level investment choices.

Both regressions account for industry and year fixed-effects as well as age controls. Industry refers to 2-

digit level detail according to NACE Rev. 2, covering the non-farm non-financial business sector (industry

codes 10-83, excluding 64-66), year to the period 2008-13. All regressions are clustered at the firm level.

Robust standard errors are reported in parenthesis. *** denotes statistical significance at the 1% level, **

significance at the 5% level, and * and the 10% level.

Source: Calculations based on Dealogic and Thomson Reuters Worldscope.

Table 5. Robustness: controlling for confounding factors

Panel A: Controlling for age and size effects

Dependent variable: capital expenditures

I II III

Fixed effects Industry Industry; Year Industry; Year;

Industry-Year

L.MFP 0.0955*** 0.0950*** 0.1002*** (0.0098) (0.0098) (0.0099)

HY shock -2.1442*** (0.1418)

L.MFP#HY shock -0.4921*** -0.5009*** -1.2501*** (0.0898) (0.0905) (0.2740)

L.Q1 0.0032 0.0034 0.0032 (0.0022) (0.0022) (0.0021)

Sales/Capital 0.0041*** 0.0041*** 0.0041*** (0.0004) (0.0004) (0.0004)

L. Employees(log) -0.0201*** -0.0202*** -0.0202*** (0.0029) (0.0029) (0.0029)

Age -0.0123*** -0.0124*** -0.0121*** (0.0010) (0.0010) (0.0010)

│ 31

Observations 21,632 21,632 21,626

R-squared 0.1662 0.1685 0.1777

Adj. R2 0.164 0.166 0.166

Panel B: Controlling for macroeconomic cycles with the output gap

I II III

Fixed effects Industry Industry; Year Industry; Year;

Industry-Year

L.MFP 0.1117*** 0.1078*** 0.1196*** (0.0128) (0.0126) (0.0231)

HY shock -1.5885***

(0.1381)

L.MFP#HY shock -0.4650*** -0.4713*** -1.2244*** (0.0937) (0.0943) (0.2814)

L.MFP#Output gap 0.0006 -0.0002 0.0018 (0.0023) (0.0022) (0.0066)

L.Q1 0.0070*** 0.0071*** 0.0069*** (0.0021) (0.0021) (0.0021)

Sales/Capital 0.0046*** 0.0046*** 0.0045*** (0.0004) (0.0004) (0.0004)

Observations 21,632 21,632 21,626

R-squared 0.1453 0.1476 0.1576

Adj. R2 0.143 0.145 0.145

Panel C: Controlling for confounding firm-level factors

Dependent variable: capital expenditures I II III

Fixed effects Industry Industry; Year Industry;

Year;

L.MFP 0.1088*** 0.1078*** 0.1115***

(0.0103) (0.0104) (0.0105)

HY shock -0.9935***

(0.1680)

L.MFP#HY shock -0.2016** -0.2038** -0.6851**

(0.0887) (0.0894) (0.2845)

L.Q1 0.0092*** 0.0093*** 0.0086***

(0.0023) (0.0023) (0.0023)

L.Q1#HY shock -0.2598*** -0.2599*** -0.2146***

(0.0667) (0.0671) (0.0668)

Sales/Capital 0.0046*** 0.0046*** 0.0046***

(0.0004) (0.0004) (0.0004)

Sales/Capital# HY shock -0.0153 -0.0155 -0.0119

(0.0107) (0.0107) (0.0113)

Observations 21,632 21,632 21,626

R-squared 0.1500 0.1523 0.1604

Adj. R2 0.148 0.150 0.148

32 │

Note: This table reports robustness checks of the baseline regression, where Panel A further controls for age

and size effects; Panel B accounts for possible confounding macroeconomic effects by interacting MFP

with the output gap; and Panel C tests if the HY shock could still pick up confounding firm-level factors by

accounting for its interaction with all other variables. Industry refers to 2-digit level detail according to

NACE Rev. 2, covering the non-farm non-financial business sector (industry codes 10-83, excluding 64-

66), year to the period 2008-13. All regressions are clustered at the firm level. Robust standard errors are

reported in parenthesis. *** denotes statistical significance at the 1% level, ** significance at the 5% level,

and * and the 10% level.

Source: Calculations based on Dealogic and Worldscope.

6.4. Discussion and implications for further research

72. Our results provide evidence that expansionary monetary policy changes the

composition of risk in primary corporate bond markets through a risk-taking channel.

Specifically, proxy SVAR results imply that HY bond issuance increases relative to IG

bond issuance. At the same time, the spread between HY bond yields and IG bond yields

decreases which points towards an increased appetite for risk of investors. Using the

estimated changes in the HY shock due to shocks to unconventional monetary policy

interacted with MFP in an otherwise standard firm-level reallocation model, we provide

evidence that this change in the composition of risk in corporate bond markets tends to

be associated with less allocative efficiency, i.e. capital does not seem to flow towards

its most productive use as strongly as in the absence of UMP shocks. The firm-level

results are robust to different fixed effect structures, the inclusion of age and firm size

as additional controls, interactions of the HY shock with other variables and accounting

for selection bias induced by UMP shocks themselves.

73. We discuss three such possibilities: (i) using proceeds for non-productive

purposes, (ii) using proceeds for intangible investment, and (iii) using proceeds to build

cash-holdings. However, it is important to bear in mind that our analysis focusses on a

specific segment of the financial markets, namely the market for corporate bonds, and a

specific firm-level outcome, namely investment in tangible capital. Therefore, a

potential mechanism underlying our results is that firms which issue bonds actually use

the proceeds for a different purpose than investment in tangible capital.

Use of proceeds

74. With a view to better understanding this heterogeneous effect of HY shocks on

investment behaviour across firms with different levels of MFP, we inspect the firms’

use of proceeds, by exploiting the idea that inherent firm-characteristics could interact

with firm-level preferences for particular uses of bond revenues. One possibility is that

low MFP firms have a higher propensity to divert corporate bond proceeds to productive

investments than their high-productive peers, who use the capital to pay dividends to

their investors, or re-finance outstanding debt at a lower rate of interest, at the expense

of capital investments. For instance, Çelik et al. (2015) show that the global share of

funds raised through bond issuance that are allocated to investment fell sharply since

2008, alongside with an increase in the share of funds used for debt reduction and

refinancing purposes.

75. The Dealogic database provides some background information on the use of the

proceeds from HY bond income by firms. However, after matching the Dealogic and

Worldscope database Figure 8 displays no meaningful difference in the share of bonds

attributed to the various categories of uses across quartiles of MFP. This may not be

surprising as the data comes with a number of caveats.

│ 33

firms are not obliged to disclose the exact use of their proceeds, and more

granular data would be warranted to accurately disentangle the different

categories firms can chose from;

the category “general corporate purposes” potentially contains a variety of uses

and could disguise important heterogeneities across firms’ uses of proceeds

from bond issuance;

for the purpose of Dealogic, firms can attribute each bond to several categories

without providing information on its predominant use.

Figure 8. The use of HY corporate bonds by MFP quartile

Share of total HY bond issuance

Note: This graph represents the share of the total HY bond issuance used for various purposes as indicated

by the issuer in the Dealogic database over the period 2008-16. Where bonds are attributed to more than

one purpose, each category is weighted equally. Within the available categories for the use of proceeds, we

considered the following ones: capital expenditures (investment), dividend payments to investors, research

and development (R&D), working capital, acquisitions and leveraged buyouts, general corporate purposes,

the repayment of debt and refinancing purposes.

Source: Calculations based on Thomson Reuters Worldscope and Dealogic.

Intangible capital

76. Proceeds could be used to invest in intangible capital, which are not included in

our analysis due to data limitations. In many countries, growth of intangible capital has

now surpassed that of tangible investment (Corrado and Hulten, 2010; Corrado et al.

2012, 2016). Recent evidence (Crouzet and Eberly, 2018a, 2018b; Gutiérrez and

Philippon, 2017b) points to a growing role of intangibles– such as R&D, design and new

business processes – in explaining weak fixed capital accumulation over the past two

decades. This change is also reflected in the changing capital composition of firms

(Corrado, Hulten and Sichel, 2005).

77. Intangibles are commonly considered a key driver of productivity (Andrews and

De Serres, 2012; Andrews and Criscuolo, 2013), especially since their contribution to

MFP growth appears to be stronger than for tangible assets (OECD, 2015b). This is

0

20

40

60

80

100

1 2 3 4

Capital expenditure Leveraged Buyout

Dividends General Corporate Purpose

R&D Use Repay Debt

Working Capital Refinancing

Acquisition

34 │

related to the non-rivalrous nature of intangibles: costs incurred at the initial stage of

developing certain types of knowledge are not re-incurred when that knowledge is used

again (Andrews and Criscuolo, 2013). Moreover, intangible investment can lead to

increasing returns to scale, as well as create knowledge that spills over into other parts

of the economy (OECD, 2015b).

78. Against this background, it is possible that the additional access to finance

granted by unconventional monetary policy may have induced firms, especially those at

the upper range of the productivity distribution, to partially substitute their tangible

capital investment for more productive intangible assets. This could then explain why

low productive firms that have less scope for absorbing intangible assets increase

tangible investment in reaction to UMP-induced HY shocks, while high productive firms

replace tangible for intangible capital. Whether such a “capital substitution effect”

occurred and was sizable enough to offset the decline in tangible investment, however,

remains speculative so long as data constraints limit further analysis.

Cash holdings

79. Another hypothesis is that high MFP firms may have reacted to HY shocks by

piling up more cash. Considering the economic uncertainty characterising the crisis

years, precautionary motives may have led firms to increase their cash holdings to ensure

greater flexibility. This would be beneficial if firms either face uncertainty about future

transactions, or tight credit constraints (Sànchez and Yurdagul, 2013). For instance,

firms spending significant amounts of resources on innovation may find it harder to

borrow at any point in time, in which case cash holdings are a necessary backup solution.

In this regard, it is no coincidence that the rise of intangible assets has been identified as

the fundamental driver of the upwards trend in US corporate cash holdings over the past

decades (Falato and Sim, 2013).

80. A simple test of whether expansionary unconventional monetary policy drove

the increase in cash holdings is to replace the left hand side variable of our baseline

equation, capital expenditures, with a variable of cash holdings. However, the results

show no sign of correlation between HY shocks and the cash holdings variable (Table

6).

Table 6. Testing for the use of proceeds with balance-sheet data

Dependent variable: Cash Industry FE Industry,

Year FE

Industry, Year,

Industry-Year FE

L.MFP 0.2920* 0.2918* 0.2903** (0.1702) (0.1710) (0.1468)

HY shock 25.4529

(19.6851)

L.MFP#HY shock -3.0234 -3.0511 -10.8141 (2.2568) (2.3035) (7.6959)

L.Q1 -0.0332** -0.0333** -0.0307** (0.0166) (0.0164) (0.0123)

Sales/Capital 0.0070* 0.0071* 0.0063** (0.0036) (0.0037) (0.0029)

Observations 15,723 15,723 15,714

R-squared 0.0218 0.0220 0.0889

Adj. R2 0.0188 0.0187 0.0708

│ 35

Note: The dependent variable, cash holdings, is scaled by total assets. Industry refers to 2-digit level detail

according to NACE Rev. 2, covering the non-farm non-financial business sector (industry codes 10-83,

excluding 64-66); year to the period of 2008-13. All regressions are clustered at the firm level. Robust

standard errors are reported in parenthesis. *** denotes statistical significance at the 1% level, **

significance at the 5% level, and * and the 10% level.

Source: Calculations based on Thomson Reuters Worldscope and Dealogic.

81. Taken together, we thus find no evidence that the use of proceeds is driving the

decline in investment observed for high productive firms in Figure 6. In view of the data

constraints faced in this analysis, however, further research is warranted to explore the

estimated heterogeneity across the productivity distribution in response to a rise in HY

bond issuance due to UMPs. In particular, investment in intangible capital seems to be

a promising avenue for further research. Thus without further empirical evidence on the

effect of shifts in the composition of risk in corporate bond markets due to expansionary

unconventional monetary policy on investment in intangibles, we refrain from

extrapolating our firm-level investment results to an aggregate level. This is because the

investment in intangibles might on aggregate balance the negative effects on allocative

efficiency found for investment in tangible capital.

7. Conclusion and policy implications

7.1. Conclusion

82. The aim of this paper is to shed light on the relationship between expansionary

monetary policy, access to high-yield bond markets and productivity in the US between

2008 and 2016.

83. Our first contribution is to provide empirical evidence that expansionary

unconventional monetary policy induces a shift in the composition of risk on corporate

bond markets by increasing the issuance of HY bonds relative to IG bonds. We further

show that this shift in risk results from higher investor demand for HY bonds, i.e.

through a risk-taking channel activated by monetary policy. Our estimates point toward

reductions in the perceived default risk as the main driver of this additional appetite for

risk of institutional investors. The subsequent investors´ search for yield in the high-

yield segment of the bond market contributes to increase the access to finance of riskier

firms.

84. The second contribution of this work is to assess the effect of the higher HY

bond issuance induced by monetary policy on the efficiency of capital reallocation. We

therefore examine whether more productive firms succeed in attracting more capital than

less productive firms, suggesting a deterioration in allocative efficiency. Our results

show that risk-taking induced by monetary policy easing disproportionally favoured

investment in tangible assets in low productive firms. We provide tentative evidence

that this finding could potentially result from a crowding-out effect. However, due to

data limitations our analysis is confined to one segment of the financial market (the bond

market) and one specific firm-level outcome (tangible investment). We therefore discuss

whether firms might allocate the proceeds of bond issuance to different purposes, most

notably investment in intangible capital. If so, results might mask a “capital substitution

effect”, where the additional access to finance granted by unconventional monetary

policy may have incentivized the most productive firms to partially substitute their

tangible capital investment for intangibles. In the event of such a substitution,

productivity gains from intangible investment may have offset, in the aggregate, the

productivity losses due to the decline in tangible investment. In the absence of further

36 │

analysis exploring this angle, however, the discussion remains speculative for the time

being.

85. Alternatively, we also assess whether high productive firms may have used

corporate bond proceeds for reasons other than productive investment (e.g. paying

dividends to their investors or building up cash holdings), thereby reducing the overall

amount of capital invested. However, we do not find any evidence supporting this

hypothesis, with the data at hand. Further research using more granular data on the use

of proceeds would be needed to disentangle this and additional mechanisms such as debt

refinancing and financial structure.

7.2. Policy implications

86. Our results suggest that, while UMP interventions have been key for dragging

the US economy out of the Great Financial Crisis, their ex post assessment points to side

effects that need to be taken into account in future monetary and financial policy

conduct.

87. First of all, UMPs have increased access to the corporate bond market for risky

firms as well as investors’ appetite for risk, changing the risk structure of this market.

Together with the well-documented substitution of bank loans with corporate bond

issuance (see Lhuissier et al., 2018) induced by the tighter regulatory environment after

the crisis, this change has shifted financial market risk from banks to holders of corporate

bonds, mainly large institutional investors like pension funds, insurance companies, and

a wide range of hedge funds and other managed accounts. In other words, risk has

moved from a currently highly regulated to a probably less regulated segment of the US

financial market. With signs that the overall level of riskiness has increased since the

crisis, this move could foreshadow further financial turbulences in the future. Therefore,

financial market monitoring could benefit from supplementing traditional financial

indicators, which largely focus on banks, with additional indicators for instance sourced

from balance sheets of institutional investors.

88. Second, the changes in access to and appetite for risky financial instruments

triggered by UMPs can have consequences for the real economy via their effects on the

allocation of capital and aggregate productivity growth. While more work is needed to

firmly establish the overall direction of these effects, the preliminary evidence produced

in this paper points to a tendency to misallocate tangible investment towards less

productive uses, with potentially negative consequences for aggregate productivity. To

an extent, this matches the short-term aim of monetary policy to sustain firms and

economic activity during the recession, but undoing the effects of the resulting

misallocation on productivity in the longer-term may prove difficult, pointing to the

need for policy to address this potential trade off. Further research should ascertain

whether our preliminary results are robust to covering a wider set of financial

instruments and to adopting a broader definition of capital including intangibles.

However, if our results prove robust, they would suggest the need to flank extraordinary

monetary policy expansions with measures in financial, product and labour markets that

enhance the ability of the economy to efficiently allocate capital.

89. Finally, our analysis only focused on one country and UMP episodes, but it

would be interesting to verify whether similar evidence would result from extending it

to more countries and different types of monetary policies, including periods of both

unconventional and conventional monetary easing.

│ 37

References

Acharya, V. and Z. Xu (2017), Financial dependence and innovation: The case of public versus private

firms, Journal of Financial Economics 124 (2), 223–243.

Adalet McGowan, M., D. Andrews and V. Millot (2017a), “The Walking Dead? Zombie Firms and

Productivity Performance in OECD countries”, OECD Economics Department Working Paper,

No 1372.

Adalet McGowan, M., D. Andrews and V. Millot (2017b), "Insolvency Regimes, Technology Diffusion

and Productivity Growth: Evidence from Firms in OECD Countries", OECD Economics

Department Working Papers, No. 1425, Paris. http://dx.doi.org/10.1787/36600267-en

Adalet McGowan. M. and D. Andrews (2015a), “Labour market mismatch and labour productivity:

Evidence from PIAAC”, OECD Economics Department Working Papers, No 1209.

Adalet McGowan, M. and D. Andrews (2015b), “Skill Mismatch and Public Policy in OECD

Countries”, OECD Economics Department Working Papers, No. 1210, Paris.

http://dx.doi.org/10.1787/5js1pzw9lnwk-en

Aghion, P., A.Bergeaud, G. Cette, R. Lecat, H, Maghin (2018) “The Inverted-U Relationship Between

Credit Access and Productivity Growth”.

Aghion, P., and R. Griffith (2005), Competition and growth, MIT Press.

Andrews, D. and F. Petroulakis (2017), "Breaking the Shackles: Zombie Firms, Weak Banks and

Depressed Restructuring in Europe", OECD Economics Department Working Papers, No.

1433, OECD Publishing, Paris, https://doi.org/10.1787/0815ce0c-en

Andrews, D. and F. Cingano (2014), “Public Policy and Resource Allocation: Evidence from Firms in

OECD Countries”, Economic Policy, No. 29(78), pp. 253-296.

Andrews, D. and C. Criscuolo (2013), “Knowledge-Based Capital, Innovation and Resource

Allocation”, OECD Economics Department Working Papers, No. 1046, Paris.

http://dx.doi.org/10.1787/5k46bj546kzs-en

Andrews, D. and A. de Serres (2012), “Intangible Assets, Resource Allocation and Growth: A

Framework for Analysis”, OECD Economics Department Working Papers, No. 989, OECD

Publishing, Paris. http://dx.doi.org/10.1787/5k92s63w14wb-en

Bakhtiari, S. (2017), “Corporate credit ratings: Selection on size or productivity?”, International Review

of Economics & Finance, 49, issue C, p. 84-101.

Bekaert, G., M. Hoerova and M. Lo Duca (2013), “Risk, Uncertainty and Monetary Policies”, Journal

of Monetary Economics, Volue 60, Issue 7.

Bloom, N. (2007), “Uncertainty and the Dynamics of R&D”, SIEPR Discussion Paper No. 07-21.

http://www-siepr.stanford.edu/RePEc/sip/07-021.pdf

Buera, F. J., J. P. Kaboski, and Y. Shin (2011):”Finance and Development: A Tale of Two Sectors”,

American Economic Review, pp. 1964-2002.

Borio, C. (2018) Remarks at the Joint BIS-IMF-OECD Conference on Weak Productivity: The Role of

Financial Factors and Policies, co-sponsored by the Global Forum on Productivity .

38 │

Borio, C., E. Kharroubi, C. Upper and F. Zampolli (2016), “Labour reallocation and productivity

dynamics: financial causes, real consequences”, BIS Working Papers, No 534, January 2016.

Borio, C. and H. Zhu (2008):”Capital regulation, risk-taking and monetary policy: a missing link in the

transmission mechanism”, BIS Working Papers, No 268.

Brown, J. R., S. M. Fazzari, and B. C. Petersen (2009), “Financing innovation and growth: Cash flow,

external equity, and the 1990s r&d boom”, The Journal of Finance 64 (1), 151– 185.

Byrne, D. M., J . G. Fernald and M. B. Reinsdorf (2016):”Does the United States Have a Productivity

Slowdown or a Measurement Problem”, Brookings Paper on Economic Activity, Spring 2016.

Caldera Sánchez, A. and D. Andrews (2011), “To Move or not to Move: What Drives Residential

Mobility Rates in the OECD?”, OECD Economics Department Working Paper Series, No. 846,

OECD Publishing.

Çelik, S., G. Demirtaş and M. Isaksson (2015), "Corporate Bonds, Bondholders and Corporate

Governance", OECD Corporate Governance Working Papers, No. 16, OECD Publishing,

Paris,https://doi.org/10.1787/5js69lj4hvnw-en.

Cette, G., J. Fernald and B. Mojon (2016), “The pre-Great Recession slowdown in productivity,”

European Economic Review, pp. 1-120.

Cihak, M., A. Demirguc-Kunt, E. Feyen and R. Levine (2013):”Financial Development in 205

Economies, 1960 to 2010”, Journal of Financial Perspectives, Vol 1(2).

Corrado, C. A., J. Haskel, M. Iommi, and C. Jona Lasinio (2012), Intangible capital and growth in

advanced economies: Measurement and comparative results.

Corrado, C. A. and C. R. Hulten (2010), “How do you measure a ”technological revolution”?” American

Economic Review 100 (2), 99–104.

Corrado, C., C. Hulten, and D. Sichel, 2005, “Measuring capital and technology: an expanded

framework” in “Measuring capital in the new economy”. University of Chicago Press.

Crouzet, N. and J. Eberly, 2018a, “Understanding Weak Capital Investment: the Role of Market

Concentration and Intangibles”, Jackson Hole Economic Policy Symposium Proceedings.

Crouzet, N. and J. Eberly, 2018b, “Intangibles, investment, and efficiency”, AEA Papers and

Proceedings.

Decker, R., J. Haltiwanger, R. Jarmin and J. Miranda (2016), “Changing Business Dynamism: Volatility

of Shocks vs. Responsiveness to Shocks”, mimeo.

Demirguc-Kunt, A. and V. Maksimovic (2002):”Funding growth in bank-based and market-based

financial systems: evidence from frim-level data”, Journal of Financial Economics, Vol. 65(3).

Demmou, L., I. Stefanescu and A. Arquié (2018), “Labour productivity growth and finance: The role of

intangible assets", ECO/CPE/WP1(2018)21.

Dlugosch, D. and T. Koluk (2017), "Energy prices, environmental policies and investment: Evidence

from listed firms", OECD Economics Department Working Papers, No. 1378, OECD

Publishing, Paris,https://doi.org/10.1787/ef6c01c6-en.

Duval, R., G. H. Hong, and Y. Timmer (2017), “Financial Frictions and the Great Productivity

Slowdown”, IMF Working Paper Series, No. 17/129.

Falato, A., D. Kadyrzhanova, and J. Sim (2013), Rising intangible capital, shrinking debt capacity, and

the US corporate savings glut.

│ 39

Fawley, B. and C. Neely (2013), “Four Stories of Quantitative Easing”, Federal Bank of St Louis, Issue

January, pp. 51-88.

Foley-Fisher, N., and R. Ramcharan, E. Yu (2016), “The impact of unconventional monetary policy on

firm financing constraints: Evidence from the maturity extension program”, Journal of

Financial Economics, Vol. 122.

Foster, L., C. Grim and J. Haltiwanger (2016), “Reallocation in the Great Recession: Cleansing or

Not?”, Journal of Labour Economics, Vol. 34.

Gal, P. (2013), "Measuring Total Factor Productivity at the Firm Level using OECD-ORBIS", OECD

Economics Department Working Papers, No. 1049, OECD Publishing, Paris,

https://doi.org/10.1787/5k46dsb25ls6-en.

Gala, V., and B. Julio (2016), “Firm Size and Corporate Investment” Retrieved

from https://repository.upenn.edu/fnce_papers/30

Gilchrist, S. and E. Zakrajsek (2013), “The Impact of the Federal Reserve’s Large Scale Asset Purchase

Programs on Corporate Credit Risk”, NBER Working Paper, No. 19337.

Gertler, M., and P. Karadi (2015), “Monetary Policy Surprises, Credit Costs, and Economic Activity“,

American Economic Journal: Macroeconomics, Vol. 7(1).

Gopinath, G., Kalemli-Özcan, S., Karabarbounis, L., and C. Villegas-Sanchez (2018). "Capital

Allocation and Productivity in South Europe*," The Quarterly Journal of Economics, vol 132

(4), pages 1915-1967

Gilchrist, S., and E. Zakrajsek (2012), “Credit Spreads and Business Cycle Fluctuations”, American

Economic Review, Vol. 102(4).

Gordon, R. (2012), “Is the US economic growth over? Faltering innovation confronts the six

headwinds”, NBER Working Paper, No. 18315

Gutierrez, G. and T. Philippon (2016), “Investment-less growth: An empirical investigation.

Technical report”, National Bureau of Economic Research.

Hau, H. and S. Lai (2016), “Asset Allocation and monetary policy, evidence from the Eurozone”,

Journal of Financial Economics, pp. 309-329.

Heckman, J. J. (1979), “Sample selection bias as a specification error”, Econometrics, Vol. 47, No.1,

Jan. 1979, pp. 153-161.

Heil, M. (2017):”Finance and Productivity: A literature review”, OECD Department Working Papers,

No. 1374.

Hsieh, C. T. and P. J. Klenow, (2017), "The Reallocation Myth," in Federal Reserve Bank of Kansas

City Economic Symposium on "Fostering a Dynamic Market Economy”.

Hsu, P.-H., X. Tian, and Y. Xu (2014), Financial development and innovation: Cross-country evidence.

Journal of Financial Economics 112 (1), 116–135.

IMF (2015), World Economic Outlook, Chapter 4, IMF Publishing.

Ioannidou, V., S. Ongena and J. Peydro (2015), “Monetary Policy, Risk-Taking, and Pricing: Evidence

from a Quasi-Natural Experiment”, Review of Finance, pp. 95-144.

Jiménez, G. , Ongena, S. , Peydró, J. and Saurina, J. (2014), “Hazardous Times for Monetary Policy:

What Do Twenty‐Three Million Bank Loans Say About the Effects of Monetary Policy on

Credit Risk‐Taking?”, Econometrica, 82: 463-505.

40 │

Kang, J., S. and S. Piao (2015), “Production Offshoring and Investment by Japanese Firms”, IMF

Working Paper WP/15/183.

Kortum, S. and J. Lerner (1998), “Does Venture Capital Spur Innovation”, NBER Working Paper No.

6846.

Krishnamurthy and Vissing-Jorgensen (2011):”The Effects of Quantitative Easing on Interest Rates:

Channels and Implications for Policy”, Brookings Papers on Economic Activity, 2011 No. 2.

Lhuissier, S., and U. Szczerbowicz (2017), “Corporate Debt Structure and Unconventional Monetary

Policy in the United States”, Working Paper.

Madsen, J. B. and J. B. Ang (2014),”Finance-led growth in the OECD since the nineteenth century:

How does financial development transmit to growth”, Review of Economics and Statistics, pp.

552-572.

Mertens, K., and M. O. Ravn (2013), “The Dyanmic Effects of Personal and Corporate Income Tax

Changes in the United States”, American Economic Review, Vol. 103(4).

Morais, B., J. L. Peydro and C. Ruiz (2015), “The International Bank Lending Channel of Monetary

Policy Rates and QE: Credit Supply, Reach-for Yield, and Real Effects”, International Finance

Discussion Papers– Board of Governors of the Federal Reserve System No. 1137.

Obstfeld, M. (2018) Remarks at the Joint BIS-IMF-OECD Conference on Weak Productivity: The Role

of Financial Factors and Policies, co-sponsored by the Global Forum on Productivity

OECD (2015a), “Growth companies, access to capital markets and corporate governance”, OECD

Report to G20 Finance Ministers and Central Bank Governors.

OECD (2015b), The Future of Productivity, OECD Publishing.

Rajan (2006), “Has finance made the world riskier?”, European Financial Management, pp. 499-533.

Reis, R. (2013), “The Portugese Slump and Crash and the Euro Crisis”, Brookings Papers on Economic

Activity, pp. 143-193.

Restuccia, D., R. Rogerson, (2008), "Policy Distortions and Aggregate Productivity with Heterogeneous

Plants," Review of Economic Dynamics, Elsevier for the Society for Economic Dynamics, vol.

11(4), pages 707-720, October.

Sànchez, J. M. and E. Yurdagul (2013), “Why are corporations holding so much cash?”, The Regional

Economist, St Louis Fed.

Stock, J.H., J.H. Wright, and M. Yogo (2002), “A Survey of Weak Instruments and Weak Identification

in Generalized Method of Moments”, Journal of Business & Economic Statistics, Vol. 20(4).

Summers, L. H. (2016),"The Age of Secular Stagnation”, Foreign Affairs, Mar/Apr 2016, 2-9.

Wooldridge, J.M. (2002), “Econometric analysis of cross section and panel data”, Cambridge, MA:MIT

Press.

│ 41

Annex A. Data

Table A.1. Summary statistic

Observations Mean Standard deviation 5% 25% 50% 75% 95%

MFP 18789 9.303 1.547 5.875 8.656 9.516 10.295 11.296

HY/IG ratio 21632 0.003 0.031 -0.046 -0.001 -0.001 0.015 0.058

Capital expenditures 21632 0.374 0.680 0.001 0.096 0.191 0.373 1.214

Tobin's Q 18965 1.579 5.795 0.000 0.007 0.066 0.417 7.135

Sales/Capital 21632 17.798 34.792 0.110 2.387 6.459 15.467 80.812

Employees 21611 6.494 2.634 1.792 4.605 6.731 8.517 10.588

Age 21632 12.161 6.121 2.000 7.000 13.000 17.000 21.000

Cash/Capital 15654 0.314 5.467 0.001 0.028 0.098 0.229 0.705

Note: This table reports the number of observations, mean, standard deviation, and various percentiles of

the variables used in our analysis. Variables are winsorized at the 1% level to reduce the effects of outliers.

The data description in the section 3 provides more detail on the variable constructions.

Source: Based on Thomson Reuters Worldscope and Dealogic.

Cleaning and matching

Filtering and Cleaning

1. Dealogic data are clean according to the following filters. First, based on the

Deal Nationality, we retain all the deals issued in US that are classified either as high

yield or investment grade. Second, we retain deals for which either the parent or the

issuer is a non-financial private company in US, namely either (i) the parent and the

issuer are private US non-financial companies, (ii) the parent is a US financial private

company while the issuer is a non-financial private company or vice-versa, (iii) the

parent is non-US but the issuer is a US non-financial private company, or (iv) the parent

is a US non-financial private company and the issuer is a non-US private financial

company. Third, to identify only manufacturing and non-financial services companies,

whether this are parent or issuer, we applied an additional filter on the sectors based on

the SIG code (Specific Industry Group code) transformed into two-digit SIC code

(Standard Industry Classification code) (see Table A2). Regarding the rating categories,

we scored them into numerical values from 1 to 21 (AAA being the highest quality rating

and CC the lowest) and we cleaned for unrealistic data in the cases where a rating

between AAA and BBB- was associated with corporate bond high yield and where a

rating between BB+ and CC was associated with corporate bond investment grade.

Finally, we consider the following categories of the variable use of proceeds which

provides the intended use of the capital raised: capital expenditures (investment),

dividend payments to investors, research and development (R&D), working capital,

acquisitions and leveraged buyouts, general corporate purposes, the repayment of debt

and refinancing purposes.

2. Worldscope data are cleaned and benchmarked using a number of common

procedures before the data are used for economic analysis (cf. Kozluk and Dlugosch,

2017 ; etc). To begin with, we correct for reporting errors by dropping values below zero

where the variable cannot take on zero values from an accounting perspective (as would

be the case for total assets), and winsorize all variables at the 1% and 99% level to

eliminate outliers that could bias the estimation results. We then apply a number of filters

to clean for double listings and non-company accounts. Lastly, all monetary variables

42 │

are deflated using 2-digit industry deflators constructed based on OECD STAN and

national accounts.

Table A.2. 2-Digit SIC (Standard Industrial Classification) Codes included in the dataset

SIC code

Name

Construction

15 General Building Contractors

16 Heavy Construction, Except Building

17 Special Trade Contractors

Manufacturing

20 Food & Kindred Products

21 Tobacco Products

22 Textile Mill Products

24 Lumber & Wood Products

25 Furniture & Fixtures

26 Paper & Allied Products

27 Printing & Publishing

28 Chemical & Allied Products

29 Petroleum & Coal Products

30 Rubber & Miscellaneous Plastics Products

32 Stone, Clay, & Glass Products

33 Primary Metal Industries

34 Fabricated Metal Products

35 Industrial Machinery & Equipment

36 Electronic & Other Electric Equipment

37 Transportation Equipment

38 Instruments & Related Products

39 Miscellaneous Manufacturing Industries

Transportation & Public Utilities

40 Railroad Transportation

41 Local & Interurban Passenger Transit

42 Trucking & Warehousing

45 Transportation by Air

46 Pipelines, Except Natural Gas

48 Communications

49 Electric, Gas, & Sanitary Services

Wholesale Trade

50 Wholesale Trade – Durable Goods

51 Wholesale Trade – Nondurable Goods

Retail Trade

52 Building Materials & Gardening Supplies

53 General Merchandise Stores

54 Food Stores

55 Automotive Dealers & Service Stations

56 Apparel & Accessory Stores

57 Furniture & Home furnishings Stores

58 Eating & Drinking Places

59 Miscellaneous Retail

Finance, Insurance, & Real Estate

65 Real Estate

Services

70 Hotels & Other Lodging Places

73 Business Services

78 Motion Pictures

81 Legal Services

87 Engineering & Management Services

│ 43

Annex B. Robustness and additional results

Figure B.1. Impulse-response functions after an expansionary monetary policy shock

Model 1: 5Y Real Rates Model 2: 5Y Real Rates

Model 1: CPI Model 2: CPI

Model 1: Industrial production Model 2: Industrial production

-0.3

-0.2

-0.1

0

0.1

1 3 5 7 9 111315171921232527293133

Bas

is p

oin

ts

Months

-0.3

-0.2

-0.1

0

0.1

1 3 5 7 9 111315171921232527293133

Bas

is p

oin

tsMonths

-0.10

0.10.20.30.40.50.6

1 3 5 7 9 111315171921232527293133

Pe

rce

nt

Months

-0.2

0

0.2

0.4

0.6

0.8

1 3 5 7 9 111315171921232527293133

Pe

rce

nt

Months

-0.4-0.2

00.20.40.60.8

1

1 3 5 7 9 111315171921232527293133

Pe

rce

nt

Months

-0.4-0.2

00.20.40.60.8

1

1 3 5 7 9 111315171921232527293133

Pe

rce

nt

Months

44 │

Model 1: HY/IG Ratio Model 2: HY/IG Ratio

Model 1: HY-IG Spread Model 2: GZ’s excess bond premium

Note: This figure depicts the impulse response functions of a one-unit standard deviation expansionary

surprise in monetary policy. We identify the monetary policy shock using an instrumental variable

approach. The IRF are shown for VAR specifications with the 5Y real rates as policy indicator and

mensualized daily surprises in 3Y real rates as instruments. Both VAR models include industrial

production and CPI as further variables. The reduced-form VAR is estimated over the 2008M06-

2016M12 period. We obtained the IRFs using a wild bootstrap with 2,000 repetitions. The solid black

line shows median responses. The dashed lines the lower and upper bounds of a 90% confidence band.

-0.2

-0.1

0

0.1

0.2

0.3

1 3 5 7 9 111315171921232527293133

Bas

is p

oin

ts

Months

-0.2

-0.1

0

0.1

0.2

0.3

1 3 5 7 9 111315171921232527293133

Bas

is p

oin

ts

Months

-0.4

-0.3

-0.2

-0.1

0

0.1

0.2

1 3 5 7 9 111315171921232527293133

Pe

rce

nta

ge p

oin

ts

Months

-0.4

-0.3

-0.2

-0.1

0

0.1

0.2

1 3 5 7 9 111315171921232527293133

Pe

rce

nta

ge p

oin

ts

Months

│ 45

Figure B.2. Different VAR estimation window: Impulse-response functions after an

expansionary monetary policy shock

5Y Real rates CPI

Industrial production HY-IG Spread

HY/IG Ratio

Note: This figure depicts the impulse response functions of a one-unit standard deviation expansionary

surprise in monetary policy. We identify the monetary policy shock using an instrumental variable

approach. The IRF are shown for VAR specifications with the 5Y real rates as policy indicator and

mensualized daily surprises in 3Y real rates as instruments. Both VAR models include industrial

production and CPI as further variables. The reduced form VAR is estimated over the 1993M01-

2016M12 period. We obtained the IRFs using a wild bootstrap with 2,000 repetitions. The solid black

line shows median responses. The dashed lines the lower and upper bounds of a 90% confidence band.

-0.3

-0.2

-0.1

0

0.1

1 3 5 7 9 111315171921232527293133

Bas

is p

oin

ts

Months

-0.1

0

0.1

0.2

0.3

1 3 5 7 9 111315171921232527293133

Pe

rce

nt

Months

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1 3 5 7 9 111315171921232527293133

Pe

rce

nt

Months

-0.4

-0.3

-0.2

-0.1

0

0.1

0.2

1 3 5 7 9 111315171921232527293133

Pe

rce

nta

ge p

oin

ts

Months

-0.2

-0.1

0

0.1

0.2

1 3 5 7 9 111315171921232527293133

Bas

is p

oin

ts

Months

46 │

OECD PRODUCTIVITY WORKING PAPERS

OECD Productivity Working Papers are published on oe.cd/GFP

01. Institutions to promote pro-productivity policies: Logic and Lessons

(November 2015) by Gary Banks

02. Frontier Firms, Technology Diffusion and Public Policy: Micro Evidence from OECD Countries

(November 2015) by Dan Andrews, Chiara Criscuolo and Peter N. Gal.

03. Policies for productivity growth

(March 2015) by Chang-Tai Hsieh

04. Could Mexico become the new “China” Policy drivers of competitiveness and productivity

(July 2016) by Sean M. Dougherty and Octavio R. Escobar

05. The Best versus the Rest: The Global Productivity Slowdown, Divergence across Firms and the Role of Public Policy

(November 2016) by Dan Andrews, Chiara Criscuolo and Peter N. Gal

06. What makes cities more productive? Agglomeration economies and the role of

urban governance: Evidence from 5 OECD countries

(February 2017) by Rudiger Ahrend, Emily Farchy, Ioannis Kaplanis and

Alexander C. Lembcke

07. Pro-Productivity Institutions: Learning From National Experience

(April 2017) by Andrea Renda and Sean Dougherty

08. The impact of structural reforms on productivity: The role of the distance to the

technological frontier

(May 2017) by Gustavo Monteiro, Ana Fountoura Gouveia and Sílvia Santos

09. Product markets’ deregulation: A more productive, more efficient and more

resilient economy?

(September 2017) by Gustavo Monteiro, Ana Fountoura Gouveia and Sílvia Santos

10. Achieving New Zealand’s productivity potential

(October 2017) by Paul Conway

11. The Contribution of Multinational Enterprises to Labor Productivity: The Case of

Israel

(February 2018) by Tatiana Slobodnitsky, Lev Drucker and Assaf Geva

12. GVCs and centrality: mapping key hubs, spokes and the periphery

(February 2018) by Chiara Criscuolo and Jonathan Timmis

13. Fear the walking dead: zombie firms, spillovers and exit barriers

(June 2018) by Ana Fontoura Gouveia and Christian Osterhold

14. GVC centrality and productivity: Are hubs key to firm performance?

(June 2018) by Chiara Criscuolo and Jonathan Timmis

15. Patterns of firm level productivity in Ireland (September 2018) by Javier Papa, Luke Rehill and Brendan O’Connor

16. Productivity Spillovers from Multinational Activity to Local Firms in Ireland (November 2018)By Mattia Di Ubaldo, Martina Lawless and Iulia Siedschlag